Search

• A feature selection method based on binary dandelion algorithm is proposed. • The algorithm applies a sine-cosine operator and a restart strategy. • Mutual information and quick bit mutation improve the performance of the algorithm. • The algorithm performs well on datasets with larger dimensions. Feature selection (FS) is an important data preprocessing technology for machine learning and data mining. Metaheuristic algorithm (MH) has been widely used in feature selection because of its powerful search function. This paper presents an improved Binary Dandelion Algorithm using Sine Cosine operator and Restart strategy (SCRBDA) for feature selection. First, the sine cosine operator is used in the radius formula of the core dandelions (CD), which significantly enhances the ability of algorithm development and exploration. Secondly, the algorithm uses a restart strategy to increase its ability to get rid of local optimum. Thirdly, mutual information is used to guide the generation of some dandelions, which pays more attention to the correlation between the selected features and categories. Finally, quick bit mutation is used as the mutation strategy to improve the diversity of the population. The SCRBDA proposed in this paper was tested on 18 datasets of different sizes from UCI machine learning database. The SCRBDA was compared with 8 other classical feature selection algorithms, and the performance of the proposed algorithm was evaluated through feature subset size, classification accuracy, fitness value, and F1-score. The experimental results show that SCRBDA achieves the best performance, which has stronger feature reduction ability and achieves better overall performance on most datasets. Especially on large-scale datasets, SCRBDA can obtain extremely smaller feature subsets while maintaining much higher classification accuracy, and satisfactory F1-score. [ABSTRACT FROM AUTHOR]

Several biological research studies have shown that the number of individuals of certain species of anurans in a specific geographical region, and the evolution of this number over time, can be used as an indicator of climate change. To detect the presence of anurans, Wireless Sensor Networks (WSNs) are usually deployed with the aim of obtaining bio-acoustic information in a set covering numerous locations. However, the identification of the anuran species from a huge number of recordings usually involves an overwhelming task that has to be undertaken by expert and intelligent systems. Previous studies into this issue have proposed several classification techniques with a common approach: they all take into account the sequential characteristic of sounds by considering syllables or other kinds of vocal segments. In noisy sounds, as it is usually the case in recordings made in natural habitats, segmentation of the signal is no straightforward task and may cause low classification accuracy. To override this problem, a new non-sequential approach is proposed in this paper. It is based on considering very small pieces of sounds (frames) each of which is then classified without considering preceding or subsequent information. Up to nine frame-based classifiers are explored in this paper and their performances are compared to the most commonly used sequential classifier: the Hidden Markov Model (HMM). Additionally, for featuring the frames, many choices have been described, although the application of the Mel Frequency Cepstral Coefficients (MFCCs) has probably become the most common method. In this work, an alternative methodology is suggested: the use of a set of MPEG-7 parameters, which offers a normalized solution with a much greater semantic content. The experimental results have shown that the proposed method clearly outperforms the HMM, thereby showing the non-sequential classification of anuran sounds to be feasible. From among the algorithms tested, the decision-tree classifier has shown the best performance with an overall classification success rate of 87.30%, which is an especially striking result considering that the analyzed sounds were affected by a decidedly noisy background. [ABSTRACT FROM AUTHOR]

Rare events, especially those that could potentially negatively impact society, often require humans’ decision-making responses. Detecting rare events can be viewed as a prediction task in data mining and machine learning communities. As these events are rarely observed in daily life, the prediction task suffers from a lack of balanced data. In this paper, we provide an in depth review of rare event detection from an imbalanced learning perspective. Five hundred and seventeen related papers that have been published in the past decade were collected for the study. The initial statistics suggested that rare events detection and imbalanced learning are concerned across a wide range of research areas from management science to engineering. We reviewed all collected papers from both a technical and a practical point of view. Modeling methods discussed include techniques such as data preprocessing, classification algorithms and model evaluation. For applications, we first provide a comprehensive taxonomy of the existing application domains of imbalanced learning, and then we detail the applications for each category. Finally, some suggestions from the reviewed papers are incorporated with our experiences and judgments to offer further research directions for the imbalanced learning and rare event detection fields. [ABSTRACT FROM AUTHOR]

Real-world optimization problems typically involve multiple objectives to be optimized simultaneously under multiple constraints and with respect to several variables. While multi-objective optimization itself can be a challenging task, equally difficult is the ability to make sense of the obtained solutions. In this two-part paper, we deal with data mining methods that can be applied to extract knowledge about multi-objective optimization problems from the solutions generated during optimization. This knowledge is expected to provide deeper insights about the problem to the decision maker, in addition to assisting the optimization process in future design iterations through an expert system. The current paper surveys several existing data mining methods and classifies them by methodology and type of knowledge discovered. Most of these methods come from the domain of exploratory data analysis and can be applied to any multivariate data. We specifically look at methods that can generate explicit knowledge in a machine-usable form. A framework for knowledge-driven optimization is proposed, which involves both online and offline elements of knowledge discovery. One of the conclusions of this survey is that while there are a number of data mining methods that can deal with data involving continuous variables, only a few ad hoc methods exist that can provide explicit knowledge when the variables involved are of a discrete nature. Part B of this paper proposes new techniques that can be used with such datasets and applies them to discrete variable multi-objective problems related to production systems. [ABSTRACT FROM AUTHOR]

In the tasks of data mining and machine learning with unlabeled data, one-class support vector machines (OCSVMs) are one of the most widely used methods for anomaly detection as one class classification. However, the choice of hyperparameters, a critical step for learning effective OCSVM decision boundaries, remains open as a post analysis step and is often undecided. To tackle this issue, this paper proposes a new data-generation method using geometrical edge probability suitable for OCSVM hyperparameter selection. Our method improves the limitations of existing methods in that the geometrical edge probability enables the generation of both pseudo target data and pseudo anomaly data. In addition, the proposed method combined with bootstrapping is able to determine the distribution of anomaly data. We evaluate the proposed method for datasets with 16 different dimensions and demonstrate the performance improvement in the classification of target and anomaly data. • Proposing geometrical edge probability to determine the data proximity to the structure. • Proposing a new data-generation method of both target and anomaly data. • Proposing an automatic selection of OC-SVM hyper-parameters. [ABSTRACT FROM AUTHOR]

Technical and quantitative analysis in financial trading use mathematical and statistical tools to help investors decide on the optimum moment to initiate and close orders. While these traditional approaches have served their purpose to some extent, new techniques arising from the field of computational intelligence such as machine learning and data mining have emerged to analyse financial information. While the main financial engineering research has focused on complex computational models such as Neural Networks and Support Vector Machines, there are also simpler models that have demonstrated their usefulness in applications other than financial trading, and are worth considering to determine their advantages and inherent limitations when used as trading analysis tools. This paper analyses the role of simple machine learning models to achieve profitable trading through a series of trading simulations in the FOREX market. It assesses the performance of the models and how particular setups of the models produce systematic and consistent predictions for profitable trading. Due to the inherent complexities of financial time series the role of attribute selection, periodic retraining and training set size are discussed in order to obtain a combination of those parameters not only capable of generating positive cumulative returns for each one of the machine learning models but also to demonstrate how simple algorithms traditionally precluded from financial forecasting for trading applications presents similar performances as their more complex counterparts. The paper discusses how a combination of attributes in addition to technical indicators that has been used as inputs of the machine learning-based predictors such as price related features, seasonality features and lagged values used in classical time series analysis are used to enhance the classification capabilities that impacts directly into the final profitability. [ABSTRACT FROM AUTHOR]

Learning from continuous streams of data has been receiving an increasingly attention in the last years. Among the many challenges related to mining data streams, change detection is one topic frequently addressed. Being able to determine whether or not data characteristics are changing along time is a major concern for data stream algorithms, be it on the supervised or unsupervised scenario. The unsupervised scenario is particularly relevant due to many practical applications do not provide target labeling information. In this scenario, most of the strategies induce consecutive models over time and compare them in order to detect data changes. In this situation, model changes are assumed to be a consequence of data modifications. However, there is no guarantee this assumption is true, since those algorithms do not rely on any theoretical background to ensure that model divergences truly indicate data changes. The need for such theoretical framework has motivated this paper to propose a new stability concept to establish bounds on the learning abilities of unsupervised algorithms designed to detect changes on data streams. This stability concept, based on the surrogate data strategy from time series analysis, provides learning guarantees for online unsupervised algorithms even in case of time dependency among observations. Furthermore, we propose a new change detection algorithm that meets the requirements of this stability concept. Experimental results on different synthetical scenarios illustrate how the stability concept proposed in this paper is applied to detect changes in unsupervised data streams. [ABSTRACT FROM AUTHOR]

• As an over-sampling method, RNSO does not require minority class instance available. • The generation of artificial minority class instances only relies on majority class. • RNSO can effectively avoid the generation of noisy instances and duplicated instances. • RNS can solve imbalanced classification task without any modification of classifier. • RNSO-based approach obtains better imbalanced classification results than other ones. The learning problem from imbalanced data set poses a major challenge in data mining community. Conventional machine learning algorithms show poor performance in dealing with the classification problems of imbalanced data set since they are originally designed to work with balanced class distribution. In this paper, we propose a new over-sampling technique, which uses the real-value negative selection (RNS) procedure to generate artificial minority data with no requirement of actual minority data available. The generated minority data with rare actual minority data if available are combined with the majority data as input to a bi-class classification approach for learning. In the experiments, we demonstrate the effectiveness of RNS in avoiding the problems often encountered by the existing over-sampling methods such as the generation of noisy instances and almost duplicated instances in the same clusters. Moreover, the extensive experimental results on the different imbalanced datasets from UCI repository and real-world imbalanced datasets show that when dealing with the classification of imbalanced datasets, the proposed hybrid approach can achieve better performance in terms of both G -Mean and F -Measure evaluation metrics as compared to the other existing imbalanced dataset classification techniques. [ABSTRACT FROM AUTHOR]

• This paper focuses on missing value imputation in class imbalanced medical datasets. • Dynamic Time Warping-Based Imputation (DTWI) is proposed. • All of the data with or without missing values in the minority class can be used for imputation. • DTWI significantly outperform the other techniques when the missing rates are larger than 30%. Class imbalanced medical datasets, such as cancer prediction, contain imbalanced numbers of data in different classes leading to skewed class distribution, which makes it very difficult for a classifier to distinguish between minority (i.e. cancer) and majority (i.e. non-cancer) classes. Related studies in the literature have proposed different types of solutions for the class imbalance problem including data level, algorithmic level, and cost-sensitive learning approaches. However, none of these potential solutions have considered the issue of missing attribute values residing in the class imbalanced medical datasets, especially for the minority class. Missing value imputation is commonly used for the construction of some models where statistical or machine learning techniques are used to produce estimations to replace the missing values. However, the existing imputation methods require a certain number of observed data to produce their estimations, the major challenge for them being that the amount of observed data (with no missing values) in the minority class is very limited, or that some data are not complete. In this paper, we proposed a novel approach, namely Dynamic Time Warping-based Imputation (DTWI), to handle class imbalanced datasets with missing values. Based on the similarity measurement technique of DTW, all of the data (with or without missing values) in the minority class can be used for missing value imputation. The experimental results based on 10 different class imbalanced medical datasets show that when the missing rates in the minority classes are smaller than 30%, DTWI performs similarly to the baseline K-NN imputation method and better than the mean/mode imputation and case deletion methods. When the missing rates are larger than 30%, DTWI significantly outperform the other techniques. [ABSTRACT FROM AUTHOR]

• We identify a lack of software frameworks for broadcast radio knowledge discovery. • A novel framework for this is proposed, using MIR and data mining technologies. • We compare radio stations via a novel SOM visualisation and similarity metric. • A method for building high-quality music datasets is proposed and demonstrated. • The use of the framework is analysed in multiple research and industry contexts. Music radio data is currently underutilised in radio program management. Software tools that listen to and analyse music airplay are in many markets nonexistent, limited, or unaffordable. In this paper we present a novel knowledge discovery and visualisation framework for broadcast radio, ZeitMetric. The ZeitMetric framework uses machine learning and music information retrieval techniques to label radio audio automatically for knowledge discovery. The framework incorporates a novel music dataset collection technique (MusiGrab) to leverage online music services for ground-truth data, as well as a novel knowledge visualisation and presentation technique based on self-organizing maps (ZeitViz). The framework is compared to what little literature relating to this topic exists, and a set of requirements for a high-quality broadcast radio knowledge discovery is developed. MusiGrab specifically is compared to an existing static music information retrieval dataset and shown to offer superior results in this context. Future research directions using and extending the framework are also discussed. On acceptance of the paper, code for a use-case of the MusiGrab dataset collection technique will be released on GitHub. [ABSTRACT FROM AUTHOR]

Highlights • Relational social recommendation provides highly detailed explanations to users. • Users find these explanations to be intriguing and often surprising. • Linking people based on entity associations reveals novel connections of interest. • Recommendation relevance and serendipity may be improved given user feedbacks. Abstract This paper outlines RSR, a relational social recommendation approach applied to a social graph comprised of relational entity profiles. RSR uses information extraction and learning methods to obtain relational facts about persons of interest from the Web, and generates an associative entity-relation social network from their extracted personal profiles. As a case study, we consider the task of peer recommendation at scientific conferences. Given a social graph of scholars, RSR employs graph similarity measures to rank conference participants by their relatedness to a user. Unlike other recommender systems that perform social rankings, RSR provides the user with detailed supporting explanations in the form of relational connecting paths. In a set of user studies, we collected feedbacks from participants onsite of scientific conferences, pertaining to RSR quality of recommendations and explanations. The feedbacks indicate that users appreciate and benefit from RSR explainability features. The feedbacks further indicate on recommendation serendipity using RSR, having it recommend persons of interest who are not apriori known to the user, oftentimes exposing surprising inter-personal associations. Finally, we outline and assess potential gains in recommendation relevance and serendipity using path-based relational learning within RSR. [ABSTRACT FROM AUTHOR]

Highlights • Here Dispersion Ratio is proposed to identify splitting attribute in Decision Trees. • The proposed method works for any type of variable: categorical, nominal, continuous. • Efficient discretization method is proposed for continuous features. • Has no bias towards features with more distinct values like many other methods. • Extensive evaluation and analysis is performed on a large number of datasets. Abstract In predictive tasks like classification, Information Gain (IG) based Decision Tree is very popularly used. However, IG method has some inherent problems like its preference towards choosing attributes with higher number of distinct values as the splitting attribute in case of nominal attributes and another problem is associated with imbalanced datasets. Most of the real-world datasets have many nominal attributes, and those nominal attributes may have many number of distinct values. In this paper, we have tried to point out these characteristics of the datasets while discussing the performance of our proposed approach. Our approach is a variant of the traditional Decision Tree model and uses a new technique called Dispersion_Ratio, a modification of existing Correlation Ratio (CR) method. The whole approach is divided into two phases - firstly, the dataset is discretised using a discretization module and secondly, the preprocessed dataset is used to build a Dispersion Ratio based Decision Tree model. The proposed method does not prefer the attributes with many unique values and indifferent about class distribution. It performs better than previously proposed CR based Decision Tree (CRDT) Model since an efficient discretization module has been added with it. We have evaluated the performance of our approach on some benchmark datasets from various domains to demonstrate the effectiveness of the proposed technique and also compared our model with Information Gain, Gain Ratio and Gini Index based models. Result shows that the proposed model outperforms other models in majority of the cases that we have considered in our experiment. [ABSTRACT FROM AUTHOR]

Highlights • A novel dynamic density peaks clustering method called DPC-DLP is proposed. • The idea of k-nearest neighbors is used to compute the cut-off and local density of points. • A graph-based label propagation mechanism to distribute labels and form final clusters. • DPC_DLP can effectively assign true labels to border points located in overlapped regions. • The results of experiments reveal the effectiveness of the proposed method. Abstract Clustering is a major approach in data mining and machine learning and has been successful in many real-world applications. Density peaks clustering (DPC) is a recently published method that uses an intuitive to cluster data objects efficiently and effectively. However, DPC and most of its improvements suffer from some shortcomings to be addressed. For instance, this method only considers the global structure of data which leading to missing many clusters. The cut-off distance affects the local density values and is calculated in different ways depending on the size of the datasets, which can influence the quality of clustering. Then, the original label assignment can cause a "chain reaction" , whereby if a wrong label is assigned to a data point, and then there may be many more wrong labels subsequently assigned to the other points. In this paper, a density peaks clustering method called DPC-DLP is proposed. The proposed method employs the idea of k-nearest neighbors to compute the global cut-off parameter and the local density of each point. Moreover, the proposed method uses a graph-based label propagation to assign labels to remaining points and form final clusters. The proposed label propagation can effectively assign true labels to those of data instances which located in border and overlapped regions. The proposed method can be applied to some applications. To make the method practical for image clustering, the local structure is used to achieve low-dimensional space. In addition, proposed method considers label space correlation, to be effective in the gene expression problems. Several experiments are performed to evaluate the performance of the proposed method on both synthetic and real-world datasets. The results demonstrate that in most cases, the proposed method outperformed some state-of-the-art methods. [ABSTRACT FROM AUTHOR]

Highlights • We propose a topic detection method for Twitter using High Utility Pattern Mining. • For every chunk of tweets, the minimum utility threshold is determined dynamically. • A method using a Topic-tree is presented for postprocessing after applying HUPM. • The proposed method shows superior topic recall performance than other methods. • The proposed method has short running time suitable for real-time application. Abstract Among internet and smart device applications, Twitter has become a leading social media platform, disseminating online events occurring in the world on a real-time basis. Many studies have been conducted to identify valuable information on Twitter. Recently, Frequent Pattern Mining has been applied for topic detection on Twitter. In Frequent Pattern Mining, a topic is considered to be a group of words that appear simultaneously, however, the method only considers the frequency of words, and their utility for topic detection is not considered in the process of pattern generation. In this paper, we propose a method to detect emerging topics on Twitter based on High Utility Pattern Mining (HUPM), which takes frequency and utility into account at the same time. For a chunk of tweets by time-based windowing on the Twitter stream, we define the utility of words based on the growth rate in frequency and find groups of words with high frequency and high utility by HUPM. For post-processing to extract actual topic patterns from candidate topic patterns generated by HUPM, an efficient data structure called Topic-tree (TP-Tree) is also proposed. Experimental results demonstrated the effectiveness of the proposed method, which showed superior performance and shorter running time than other tested topic detection methods. In particular, the proposed method showed a 5% higher topic recall than the other compared methods for the three datasets used. [ABSTRACT FROM AUTHOR]

Abstract: Quite often, in order to derive meaningful insights, accounting researchers have to analyze large bodies of text. Usually, this is done manually by several human coders, which makes the process time consuming, expensive, and often neither replicable nor accurate. In an attempt to mitigate these problems, we perform a feasibility study investigating the applicability of computer-aided content analysis techniques onto the domain of accounting research. defines an algorithm’s reliability as its stability, reproducibility and accuracy. Since in computer-aided text classification, which is inherently objective and repeatable, the first two requirements, stability and reproducibility, are not an issue, this paper focuses exclusively on the third requirement, the algorithm’s accuracy. It is important to note that, although inaccurate classification results are completely worthless, it is surprising to see how few research papers actually mention the accuracy of the used classification methodology. After a survey of the available techniques, we perform an in depth analysis of the most promising one, LPU (Learning from Positive and Unlabelled), which turns out to have an F-value and accuracy of about 90%, which means that, given a random text, it has a 90% probability of classifying it correctly. [Copyright &y& Elsevier]

Abstract: Identification of transcription factor binding sites is a key task to understand gene regulation mechanism to discover gene networks and functions. Clustering approach is proved to be useful when finding such patterns residing in promoter regions of co-regulated genes. Four clustering algorithms, Self-Organizing Map, K-Means, Fuzzy C-Means and Expectation-Maximization are studied in this paper to discover motifs in datasets extracted from Saccharomyces cerevisiae, Escherichia coli, Droshophila melanogaster and Homo sapiens DNA sequences. Required modifications to clustering algorithms in order to adapt them to motif finding task are presented through the paper. Then, their motif-finding performances are discussed carefully and evaluated against a popular motif-finding method, MEME. [Copyright &y& Elsevier]

Abstract: Because of the varying and dynamic characteristics of network traffic, such as fast transfer, huge volume, shot-lived, inestimable and infinite, it is a serious challenge for network administrators to monitor network traffic in real time and judge whether the whole network works well. Currently, most of the existing techniques in this area are based on signature training, learning or matching, which may be too complicated to satisfy timely requirements. Other statistical methods including sampling, hashing or counting are all approximate methods and compute an incomplete set of results. Since the main objective of network monitoring is to discover and understand the active events that happen frequently and may influence or even ruin the total network. So in the paper we aim to use the technique of frequent pattern mining to find out these events. We first design a sliding window model to make sure the mining result novel and integrated; then, under the consideration of the distribution and fluidity of network flows, we develop a powerful class of algorithms that contains vertical re-mining algorithm, multi-pattern re-mining algorithm, fast multi-pattern capturing algorithm and fast multi-pattern capturing supplement algorithm to deal with a series of problems when applying frequent pattern mining algorithm in network traffic analysis. Finally, we develop a monitoring system to evaluate our algorithms on real traces collected from the campus network of Peking University. The results show that some given algorithms are effective enough and our system can definitely identify a lot of potentially very valuable information in time which greatly help network administrators to understand regular applications and detect network anomalies. So the research in this paper not only provides a new application area for frequent pattern mining, but also provides a new technique for network monitoring. [ABSTRACT FROM AUTHOR]

Abstract: Adaptive Hypermedia systems are becoming more important in our everyday activities and users are expecting more intelligent services from them. The key element of a generic adaptive hypermedia system is the user model. Traditional machine learning techniques used to create user models are usually too rigid to capture the inherent uncertainty of human behavior. In this context, soft computing techniques can be used to handle and process human uncertainty and to simulate human decision-making. This paper examines how soft computing techniques, including fuzzy logic, neural networks, genetic algorithms, fuzzy clustering and neuro-fuzzy systems, have been used, alone or in combination with other machine learning techniques, for user modeling from 1999 to 2004. For each technique, its main applications, limitations and future directions for user modeling are presented. The paper also presents guidelines that show which soft computing techniques should be used according to the task implemented by the application. [Copyright &y& Elsevier]

The paper proposes an incremental Gene Expression Programming classifier for mining imbalanced datasets. Imbalanced datasets are commonly encountered in real-life applications. There exist numerous algorithms, techniques, and tools which are proposed as suitable for dealing with imbalanced class distribution. Yet, none of them seems to be able to outperform all others in all possible applications. We believe that our approach can extend the available range of learners that have proven good performance in mining imbalanced data and imbalanced streams. The idea is to adapt the GEP classifier to requirements of the imbalanced data environment with reuse of the minority class instances, and application of the incremental learning paradigm. The paper offers an overview of the related work and a detailed description of the proposed incremental learner. An extensive computational experiment, based on data from the KEEL dataset repository, proves that in numerous cases the approach is competitive to other state-of-the-art learners. • A new approach to incrementally mining imbalanced datasets is proposed. • A new mechanism for the minority class data reuse is suggested. • Gene Expression Programming is used for inducing complex expression trees. • The good performance of the approach is confirmed experimentally. [ABSTRACT FROM AUTHOR]

With the increasing popularity of large scale Knowledge Graph (KG)s, many applications such as semantic analysis, search and question answering need to link entity mentions in texts to entities in KGs. Because of the polysemy problem in natural language, entity disambiguation is thus a key problem in current research. Existing disambiguation methods have considered entity prominence, context similarity and entity-entity relatedness to discriminate ambiguous entities, which are mainly working on document or paragraph level texts containing rich contextual information, and based on lexical matching for computing context similarity. When meeting short texts containing limited contextual information, such as web queries, questions and tweets, those conventional disambiguation methods are not good at handling single entity mention and measuring context similarity. In order to enhance the performance of disambiguation methods based on context similarity with such short texts, we propose SCSNED method for disambiguation based on semantic similarity between contextual words and informative words of entities in KGs. Specially, we exploit the effectiveness of both knowledge-based and corpus-based semantic similarity methods for entity disambiguation with SCSNED. Moreover, we propose a Category2Vec embedding model based on joint learning of word and category embedding, in order to compute word-category similarity for entity disambiguation. We show the effectiveness of these proposed methods with illustrative examples, and evaluate their effectiveness in a comparative experiment for entity disambiguation in real world web queries, questions and tweets. The experimental results have identified the effectiveness of different semantic similarity methods, and demonstrated the improvement of semantic similarity methods in SCSNED and Category2Vec over the conventional context similarity baseline. We further compare the proposed approaches with the state of the art entity disambiguation systems and show the performances of the proposed approaches are among the best performing systems. In addition, one important feature of the proposed approaches using semantic similarity, is the potential application on any existing KGs since they mainly use common features of entity descriptions and categories. Another contribution of the paper is an updated survey on background of entity disambiguation in KGs and semantic similarity methods. [ABSTRACT FROM AUTHOR]

Interesting itemset mining is a fundamental research problem in knowledge management and machine learning. It is intended to identify interesting relations between variables in a database using some measures of interestingness and has a number of applications, including market basket analysis, web usage mining, intrusion detection, and many others. This paper proposes a new interestingness measure, the fault-tolerant tile. That is based on two observations: (1) the length of an itemset can be as important as its frequency; (2) knowledge discovery from real-world datasets calls for fault-tolerant data mining (e.g. extracting fault-tolerant association rules, analyzing noisy datasets). Given a user-defined fault tolerance value, we are interested in finding the maximum/top-k fault-tolerant tiles. Due to the exponential search space of candidate itemsets, both problems are NP-hard. While using some monotonic property to prune search space is a common strategy for interesting itemset mining, no monotonic property is available for this problem. To tackle the challenge, we utilize the branch-and-bound search strategy to analyze the characteristics of candidate itemsets at each searching branch and estimating their bounds. Our experimental results show that our algorithms can effectively analyze real datasets and retrieve meaningful results. [ABSTRACT FROM AUTHOR]

Due to technological advancement and the explosive growth of electrically stored information, automated methods are required to aid users in maintaining and processing this huge amount of information. Experts, as well as machine learning processes on large volumes of data, are the main sources of knowledge. Knowledge extraction is an important step in framing expert and intelligent systems. However, the knowledge extraction phase is very slow or even impossible due to noise and large size of data. To enhance the productivity of machine learning algorithms, feature selection or attribute reduction plays a key role in the selection of relevant and non-redundant features to improve the performance of classifiers and interpretability of data. Many areas like machine learning, image processing, data mining, natural language processing and Bioinformatics, etc., which have high relevancy to expert and intelligent systems, are applications of feature selection. Rough set theory has been successfully applied for attribute reduction, but this theory is inadequate in the case of attribute reduction of real-valued data set as it may lose some information during the discretization process. Fuzzy and rough set theories have been combined and various attribute selection techniques were proposed, which can easily handle the real-valued data. An intuitionistic fuzzy set possesses a strong ability to represent information and better describing the uncertainty when compared to the classical fuzzy set theory as it considers positive, negative and hesitancy degree simultaneously for an object to belong to a set. This paper proposes a novel mechanism of attribute selection using tolerance-based intuitionistic fuzzy rough set theory. For this, we present tolerance-based intuitionistic fuzzy lower and upper approximations and formulate a degree of dependency of decision features over the set of conditional features. Moreover, the basic results on lower and upper approximations based on rough sets are extended for intuitionistic fuzzy rough sets and analogous results are established. In the end, the proposed algorithm is applied to an example data set and the comparison between tolerance-based fuzzy rough and intuitionistic fuzzy rough sets approaches for feature selection is presented. The proposed concept is found to be better performing in the form of selected attributes. [ABSTRACT FROM AUTHOR]

In multiword expression (MWE) recognition, there exist many studies where different learning methods are employed to decide whether given word combination is a multiword expression. The recognition methods commonly utilize a number of features that are extracted from a data source, frequently from the given text. Though the recognition methods and the features are well studied, we believe that to achieve the best possible performance with a learning method, different subsets of features should also be considered and the best performing subset must be selected. In this paper, we propose a procedure that covers the performance comparison of well-known feature selection methods to obtain the best feature subset in MWE recognition. The evaluation tests are performed on a Turkish MWE data set and the performance is measured by precision, recall and F1 values. The highest F1 value =0.731 is obtained by C4.5 classifier employing either wrapper or filtering method in feature selection. In the regarding setting(s), it is examined that the performance is increased by 1.11% compared to the setting where all features are employed in classification. Based on the experimental results, it may be stated that feature selection improves the performance of MWE recognition by eliminating the noisy/non-effective features. Moreover, it is obvious that proposed feature selection method contributes to the overall MWE recognition system by reducing the measurement and storage requirements due to the lower number of features in classification, providing a faster and more-cost effective learning model. [ABSTRACT FROM AUTHOR]

Due to the unprecedented speed and volume of generated raw data in most of applications, data stream mining has attracted a lot of attention recently. Methods for solving these problems should address challenges in this area such as infinite length, concept-drift, recurring concepts, and concept-evolution. Moreover, due to the speedy intrinsic of data streams, the time and space complexity of the methods are extremely important. This paper proposes a novel method based on multiple-kernels for classifying non-stationary data streams, which addresses the mentioned challenges with special attention to the space complexity. By learning multiple kernels and specifying the boundaries of classes in the feature (mapped) space of combined kernels, the required amount of memory will be decreased. These kernels will be updated regularly throughout the stream when the true labels of instances are received. Newly arrived instances will be classified with respect to their distance to boundaries of the previously known classes in the feature spaces. Due to the efficient memory usage, the computation time does not increase significantly through the stream. We evaluate the performance of the proposed method using a set of experiments conducted on both real and synthetic benchmark data sets. The experimental results show the superiority of the proposed method over the state-of-the-art methods in this area. [ABSTRACT FROM AUTHOR]

High utility itemset mining problem uses the notion of utilities to discover interesting and actionable patterns. Several data structures and heuristic methods have been proposed in the literature to efficiently mine high utility itemsets. This paper advances the state-of-the-art and presents HMiner, a high utility itemset mining method. HMiner utilizes a few novel ideas and presents a compact utility list and virtual hyperlink data structure for storing itemset information. It also makes use of several pruning strategies for efficiently mining high utility itemsets. The proposed ideas were evaluated on a set of benchmark sparse and dense datasets. The execution time improvements ranged from a modest thirty percent to three orders of magnitude across several benchmark datasets. The memory consumption requirements also showed up to an order of magnitude improvement over the state-of-the-art methods. In general, HMiner was found to work well in the dense regions of both sparse and dense benchmark datasets. [ABSTRACT FROM AUTHOR]

In this paper, we propose a new decision forest algorithm that builds a set of highly accurate decision trees by exploiting the strength of all non-class attributes available in a data set, unlike some existing algorithms that use a subset of the non-class attributes. At the same time to promote strong diversity, the proposed algorithm imposes penalties (disadvantageous weights) to those attributes that participated in the latest tree in order to generate the subsequent trees. Besides, some other weight-related concerns are taken into account so that the trees generated by the proposed algorithm remain individually accurate and retain strong diversity. In order to show the worthiness of the proposed algorithm, we carry out experiments on 20 well known data sets that are publicly available from the UCI Machine Learning Repository. The experimental results indicate that the proposed algorithm is effective in generating highly accurate and more balanced decision forests compared to other prominent decision forest algorithms. Accordingly, the proposed algorithm is expected to be very effective in the domain of expert and intelligent systems. [ABSTRACT FROM AUTHOR]

When applying data-mining techniques to real-world data, we often find ourselves facing observations that have no value recorded for some attributes. This can be caused by several phenomena, such as a machine’s incapability to record certain characteristics or a person refusing to answer a question in a poll. Depending on that motivation, values gone missing may follow one kind of pattern or another, or describe no regularity at all. One approach to palliate the effect of missing data on machine learning tasks is to replace the missing observations. Imputation algorithms attempt to calculate a value for a missing gap, using information associated with it, i.e., the attribute and/or other values in the same observation. While several imputation methods have been proposed in the literature, few works have addressed the question of the relationship between the type of missing data, the choice of the imputation method, and the effectiveness of classification algorithms that used the imputed data. In this paper we address the relationship among these three factors. By constructing a benchmark of hundreds of databases containing different types of missing data, and applying several imputation methods and classification algorithms, we empirically show that an interaction between imputation methods and supervised classification can be deduced. Besides, differences in terms of classification performance for the same imputation method in different missing data patterns have been found. This points to the convenience of considering the combined choice of the imputation method and the classifier algorithm according to the missing data type. [ABSTRACT FROM AUTHOR]

Anomaly (or outlier) detection is well researched objective in data mining due to its importance and inherent challenges. An outlier could be the key discovery to be made from large datasets and the insights gathered from them could be of significance in a wide variety of domains like information security, business intelligence, clinical decision support, financial monitoring etc. Recently, Support Vector Data Description (SVDD) driven approaches are shown as having good predictive accuracy. This paper proposes a novel low-complexity anomaly detection algorithm based on Support Vector Data Description (SVDD). The proposed algorithm reduces the complexity by avoiding the calculation of Lagrange multipliers of an objective function, instead locates an approximate pre-image of the SVDD sphere's center, within the input space itself. The crux of the training algorithm is a gradient descent of the primal objective function using Simultaneous Perturbation Stochastic Approximation (SPSA). Experiments using datasets obtained from UCI machine learning repository have demonstrated that the accuracies of the proposed approach are comparable while the training time is much lesser than Classical SVDD. [ABSTRACT FROM AUTHOR]

• The chaos method is introduced to quantitative evaluate the randomness of dataset. • The chaotic hourly wind speed signal is analyzed and predicted with high precision. • The framework adopted the three-stage information mining technology is proposed. • The information mining technology in hybrid prediction method is discussed. With the development of wind power which is the great substitute for traditional energy, it is worth conducting an in-depth exploration of the hourly wind speed time series which is chaotic due to the complex weather. In this paper, the hybrid intelligent framework is proposed as an integrated prediction tool with high accuracy for signal pre-processing, data prediction, and result optimization of short-term hourly wind speed. Specifically, its excellent performance is guaranteed through adequate information extraction on three stages. The first stage is completed in the signal pre-processing module that the interference information is cleaned up from the hourly wind speed signal via de-noising. The second stage is conducted in the data prediction module that the hidden regular information is fully extracted via the decomposition method. The third stage is performed in the resulting optimization module that the residual information is recovered via error modification. For illustration, the performance of the proposed framework is evaluated through historical hourly wind speed, taken from publicly available Sotavento wind farms. The obtained experimental results indicate that de-noising is beneficial to capturing the real trend, but it may negatively impact short-term prediction accuracy. For the hybrid prediction model based on the empirical mode decomposition-based method, the de-noising mode integrating into the decomposition process is more effective than independent de-noising. The second stage is the key to improving the forecasting performance that, adopting the decomposition method, the average fitting performance is improved by 52.84% than the single models. Before the third stage, chaos test is necessary to determine whether there is a requirement of error modification. In summary, the proposed prediction framework can capture the complex characteristics for different short-term hourly wind speed time series, achieving greater performance than the other comparative models. [ABSTRACT FROM AUTHOR]

• A clustering-based technique is proposed to speed-up association rule mining (ARM). • Fuzzy logic, data mining and machine learning relations are considered in the paper. • The paper contributes to the comparative analysis of fuzzy and non-fuzzy ARM. In this paper, a case study on the role of fuzzy logic (FL) in data mining and machine learning is carried out. It is outlined that, in order to draw more attention of data-mining and machine-learning communities to FL, studies on FL could be more focused not on the activities that fuzzy methods can perform better but rather on the activities that fuzzy methods can perform and the non-fuzzy ones can't. Such approach takes us away from discussing quantitative differences between fuzzy and non-fuzzy methods to discussing qualitative differences, which are possibly more favorable objects of scientific curiosity. Following the outlined suggestion, a novel speed-up technique is proposed in this paper to support association rule mining (ARM). The proposed technique is a clustering-based one and provides fusion of clustering and ARM. The catchy feature of this technique is that it works well if applied in fuzzy ARM and doesn't work well if applied in non-fuzzy ARM. The proposed technique is put through experimental verification involving several real-world datasets, and the results substantiate its effectiveness. [ABSTRACT FROM AUTHOR]

In Data Mining, during the preprocessing step, there is a considerable diversity of candidate algorithms to select important features, according to some criteria. This broad availability of algorithms that perform the Feature Selection task gives rise to the difficulty of choosing, a priori , between the algorithms at hand, the most promising one for a particular problem. In this paper, we present the proposal and evaluation of a new architecture for the recommendation of Feature Selection algorithms based on the use of Metalearning. Our framework is very flexible since the user can adapt it to its proper needs. This flexibility is one of the main advantages of our proposal over other approaches in the literature, which involve steps that cannot be adapted to the user’s local requirements. Furthermore, it combines several concepts of intelligent systems, including Machine Learning and Data Mining, with topics derived from expert systems, as user and data-driven knowledge, with meta-knowledge. This set of solutions coupled with leading-edge technologies allows our architecture to be integrated into any information system, which impact on the automation of services and in reducing human effort during the process. Regarding the Metalearning process, our framework considers several types of properties inherent to the data sets, as well as, Feature Selection algorithms based on many information, distance, dependence and consistency measures. The quality of the methods for Feature Selection was estimated according to a multicriteria performance measure, which guided the ranking process of these algorithms for the construction of data metabases. Proposed by the authors of this work, this multicriteria performance measure combines any three measurements on a single one, creating an interesting and powerful tool to evaluate not only FS algorithms but also to assess any context where it is necessary a combination to maximize a measure or minimize it. The recommendation models, represented by decision trees and induced from the training metabases, allowed us to see in what circumstances a Feature Selection algorithm outperforms the other and what aspects of the data present greater influence in determining the performance of these algorithms. Nevertheless, if the user wishes, any other learning algorithm may be used to induce the recommendation model. This versatility is another strong point of this proposal. Results show that with the characterization of data, through statistical, information and complexity measures, it is possible to reach an accuracy higher than 90%. Besides yielding recommendation models that are interpretable and robust to overfitting, the developed architecture is less computationally expensive than approaches recently proposed in the literature. [ABSTRACT FROM AUTHOR]

Recommender system is a specific type of intelligent systems, which exploits historical user ratings on items and/or auxiliary information to make recommendations on items to the users. It plays a critical role in a wide range of online shopping, e-commercial services and social networking applications. Collaborative filtering (CF) is the most popular approaches used for recommender systems, but it suffers from complete cold start (CCS) problem where no rating record are available and incomplete cold start (ICS) problem where only a small number of rating records are available for some new items or users in the system. In this paper, we propose two recommendation models to solve the CCS and ICS problems for new items, which are based on a framework of tightly coupled CF approach and deep learning neural network. A specific deep neural network SADE is used to extract the content features of the items. The state of the art CF model, timeSVD++, which models and utilizes temporal dynamics of user preferences and item features, is modified to take the content features into prediction of ratings for cold start items. Extensive experiments on a large Netflix rating dataset of movies are performed, which show that our proposed recommendation models largely outperform the baseline models for rating prediction of cold start items. The two proposed recommendation models are also evaluated and compared on ICS items, and a flexible scheme of model retraining and switching is proposed to deal with the transition of items from cold start to non-cold start status. The experiment results on Netflix movie recommendation show the tight coupling of CF approach and deep learning neural network is feasible and very effective for cold start item recommendation. The design is general and can be applied to many other recommender systems for online shopping and social networking applications. The solution of cold start item problem can largely improve user experience and trust of recommender systems, and effectively promote cold start items. [ABSTRACT FROM AUTHOR]

Feedback, without doubt, is a very important mechanism for companies or political parties to re-evaluate and improve their processes or policies. In this paper, we propose opinion influencing factors (OIFs) as a means to provide feedback about what influences the opinions of people. We also describe a methodology to mine OIFs from textual documents with the intention to bring a new perspective to the existing recommendation systems by concentrating on service providers (or policy makers) rather than customers. This new perspective enables one to discover the reasons why people like or do not like something by learning relationships among the traits/products via semantic rules and the factors that lead to change on the opinions such as from positive to negative. As a case study we target the healthcare domain, and experiment with the patients’ reviews on doctors. Experimental results show the gist of thousands of comments on particular aspects (also called as factors) associated with semantic rules in an effective way. [ABSTRACT FROM AUTHOR]

Data clustering is a very well studied problem in machine learning, data mining, and related disciplines. Most of the existing clustering methods have focused on optimizing a single clustering objective. Often, several recent disciplines such as robot team deployment, ad hoc networks, multi-agent systems, facility location, etc., need to consider multiple criteria, often conflicting, during clustering. Motivated by this, in this paper, we propose a sequential game theoretic approach for multi-objective clustering, called ClusSMOG-II. It is specially designed to optimize simultaneously intrinsically conflicting objectives, which are inter-cluster/intra-cluster inertia and connectivity. This technique has an advantage of keeping the number of clusters dynamic. The approach consists of three main steps. The first step sets initial clusters with their representatives, whereas the second step calculates the correct number of clusters by resolving a sequence of multi-objective multi-act sequential two-player games for conflict-clusters. Finally, the third step constructs homogenous clusters by resolving sequential two-player game between each cluster representative and the representative of its nearest neighbor. For each game, we define payoff functions that correspond to the model objectives. We use a methodology based on backward induction to calculate a pure Nash equilibrium for each game. Experimental results confirm the effectiveness of the proposed approach over state-of-the-art clustering algorithms. [ABSTRACT FROM AUTHOR]

In this paper we show how a vector-based word representation obtained via word2vec can help to improve the results of a document classifier based on bags of words. Both models allow obtaining numeric representations from texts, but they do it very differently. The bag of words model can represent documents by means of widely dispersed vectors in which the indices are words or groups of words. word2vec generates word level representations building vectors that are much more compact, where indices implicitly contain information about the context of word occurrences. Bags of words are very effective for document classification and in our experiments no representation using only word2vec vectors is able to improve their results. However, this does not mean that the information provided by word2vec is not useful for the classification task. When this information is used in combination with the bags of words, the results are improved, showing its complementarity and its contribution to the task. We have also performed cross-domain experiments in which word2vec has shown much more stable behavior than bag of words models. [ABSTRACT FROM AUTHOR]

Data streams classification is an important approach to get useful knowledge from massive and dynamic data. Because of concept drift, traditional data mining techniques cannot be directly applied in data streams environment. Extreme learning machine (ELM) is a single hidden layer feedforward neural network (SLFN), comparing with the traditional neural network (e.g. BP network), ELM has a faster speed, so it is very suitable for real-time data processing. In order to deal with the challenge in data streams classification, a new approach based on extreme learning machine is proposed in this paper. The approach utilizes ELMs as base classifiers and adaptively decides the number of the neurons in hidden layer, in addition, activation functions are also randomly selected from a series of functions to improve the performance of the approach. Finally, the algorithm trains a series of classifiers and the decision results for unlabeled data are made by weighted voting strategy. When the concept in data streams keeps stable, every classifier is incrementally updated by using new data; if concept drift is detected, the classifiers with weak performance will be cleared away. In the experiment, we used 7 artificial data sets and 9 real data sets from UCI repository to evaluate the performance of the proposed approach. The testing results showed, comparing with the conventional classification methods for data streams such as ELM, BP, AUE2 and Learn++.MF, on most data sets, the new approach could not only be simplest in the structure, but also get a higher and more stable accuracy with lower time consuming. [ABSTRACT FROM AUTHOR]

With the advent of Big Data, data is being collected at an unprecedented fast pace, and it needs to be processed in a short time. To deal with data streams that flow continuously, classical batch learning algorithms cannot be applied and it is necessary to employ online approaches. Online learning consists of continuously revising and refining a model by incorporating new data as they arrive, and it allows important problems such as concept drift or management of extremely high-dimensional datasets to be solved. In this paper, we present a unified pipeline for online learning which covers online discretization, feature selection and classification. Three classical methods—the k -means discretizer, the χ 2 filter and a one-layer artificial neural network—have been reimplemented to be able to tackle online data, showing promising results on both synthetic and real datasets. [ABSTRACT FROM AUTHOR]

Feature selection is a crucial preprocessing technique in data mining and machine learning and has attracted increasing attentions. However, the relevance of existing methods only contains limited information. To enhance the classification ability of filter and information theory-based feature selection methods and reduce the redundancy of the selected subset, a relevant-redundant weight-based feature criterion (FSRRW) is proposed. In this paper, a feature relevant-redundant weight (RRW) is constructed to extract the important relevant and redundant information. Then, a novel feature relevance is defined based on the weight, which contains more comprehensive information from the dynamically changing features. Additionally, a feature evaluation criterion is presented via maximizing the feature relevance and minimizing the feature redundancy. The proposed algorithm and seven compared methods are tested on 20 benchmark datasets. Extensive experiments demonstrate that the proposed criterion exhibits better feature screening abilities, effectively facilitates classification, and has preferable applicability and robustness. • Design a feature relevant-redundant weight and define feature relevance. • Construct a novel feature evaluation criterion named FSRRW. • FSRRW is evaluated on 20 benchmark datasets under three classifiers. • The proposed method outperforms seven popular feature selection methods. [ABSTRACT FROM AUTHOR]

• A data-driven product failure detection framework was developed. • Sampling strategies improved the performance of the models. • ANOVA selection technique improved the performance. • The RUSBoosted Tree algorithm provided satisfactory results. • The proposed methodology is flexible to be applied in other sectors. For a healthy production line, it is essential to ensure a low failure rate of products. Product quality in production lines can be inspected using several techniques at the end of a production process, including a manual inspection. Different methods are applied to inspect the product quality at the end of the production process and sometimes during the production. This is often done using manual inspection, but this is less efficient, expensive, and time-consuming. Machine learning algorithms have the potential for evaluating and predicting product quality in a production line. In this paper, a novel product failure detection model that applies ANOVA (Analysis of Variance) feature selection method, Min-Max Scaling normalization method, mean imputation technique, Random Forest classification algorithm, a data sampling technique, and Grid Search parameter optimization approach is proposed and validated. For the comparison of the proposed model, several experiments have been performed using five classification algorithms, including RUSBoosted Tree. Experimental results demonstrated that the proposed model using the Random Forest algorithm, ANOVA feature selection, and sampling method achieves the best performance among other models and detects the faulty products effectively. It was also shown that the RUSBoosted Tree algorithm can be considered by practitioners for building the faulty product prediction model when data sampling and feature selection techniques are not integrated into the prediction model. [ABSTRACT FROM AUTHOR]

• An overview of deep learning applications to multimodal medical data analysis. • Categorize methods into unsupervised, semi-supervised, self-supervised, supervised. • Introduce the most popular deep learning architectures in the medical field. • Introduce prevalent modalities and multimodal datasets in the medical field. • Identify common problems and open challenges in multimodal medical data analysis. Deep learning methods have achieved significant results in various fields. Due to the success of these methods, many researchers have used deep learning algorithms in medical analyses. Using multimodal data to achieve more accurate results is a successful strategy because multimodal data provide complementary information. This paper first introduces the most popular modalities, fusion strategies, and deep learning architectures. We also explain learning strategies, including transfer learning, end-to-end learning, and multitask learning. Then, we give an overview of deep learning methods for multimodal medical data analysis. We have focused on articles published over the last four years. We end with a summary of the current state-of-the-art, common problems, and directions for future research. [ABSTRACT FROM AUTHOR]

How to design an efficient method to handle mixed-attribute data classification (MADC) problems has become a hot topic in data mining and machine learning. Current MADC methods mostly transform mixed-attribute data into discrete-attribute data or continuous-attribute data before classification algorithms are trained. The discretization of continuous-attribute data usually results in information loss, while the binarization of discrete-attribute data generally yield more discrete-attributes. To address these issues, this paper proposes a novel MADC method abbreviated as DO-RVFL-NBC, which is a Dependency-Oriented aggregation model of random vector functional link (RVFL) network and naive Bayes classifier (NBC). First, the method transforms the original mixed-attribute set into a dependent attribute set and an independent attribute set by considering the variation rates of dependence and independence, respectively. Second, a RVFL network is trained based on the dependent attribute set where each attribute has a weight to represent its dependence importance degree. Third, a weighted NBC is constructed by assigning the independence importance degrees as weights for the calculation of class-conditional probability. Finally, exhaustive experiments are conducted to validate the feasibility, rationality, and effectiveness of the DO-RVFL-NBC method using 22 benchmark mixed-attribute data sets. Experimental results show that (1) dependence and independence exist in the original mixed-attribute set and can be effectively explored; (2) changes of attribute dependences can improve the generalization capabilities of the RVFL network and NBC; and (3) a statistical analysis indicates that DO-RVFL-NBC can obtain considerably better testing accuracies on the benchmark mixed-attribute data sets in comparison with 13 other MADC methods. This demonstrates that DO-RVFL-NBC is a viable approach for MADC problems. • Dependency-Oriented aggregation of RVFL network and NBC was proposed. • The new feature dependence and independence exploration strategies were designed. • The exhaustive experiments were conducted to validate the effectiveness of new model. • The reasons why model can obtain the better prediction performances were revealed. [ABSTRACT FROM AUTHOR]

Every year billions of Euros are lost worldwide due to credit card fraud. Thus, forcing financial institutions to continuously improve their fraud detection systems. In recent years, several studies have proposed the use of machine learning and data mining techniques to address this problem. However, most studies used some sort of misclassification measure to evaluate the different solutions, and do not take into account the actual financial costs associated with the fraud detection process. Moreover, when constructing a credit card fraud detection model, it is very important how to extract the right features from the transactional data. This is usually done by aggregating the transactions in order to observe the spending behavioral patterns of the customers. In this paper we expand the transaction aggregation strategy, and propose to create a new set of features based on analyzing the periodic behavior of the time of a transaction using the von Mises distribution. Then, using a real credit card fraud dataset provided by a large European card processing company, we compare state-of-the-art credit card fraud detection models, and evaluate how the different sets of features have an impact on the results. By including the proposed periodic features into the methods, the results show an average increase in savings of 13%. [ABSTRACT FROM AUTHOR]

The paper presents a new approach to the dynamic classifier selection in an ensemble by applying the best suited classifier for the particular testing sample. It is based on the area under curve (AUC) of the receiver operating characteristic (ROC) of each classifier. To allow application of different types of classifiers in an ensemble and to reduce the influence of outliers, the quantile representation of the signals is used. The quantiles divide the ordered data into essentially equal-sized data subsets providing approximately uniform distribution of [0–1] support for each data point. In this way the recognition problem is less sensitive to the outliers, scales and noise contained in the input attributes. The numerical results presented for the chosen benchmark data-mining sets and for the data-set of images representing melanoma and non-melanoma skin lesions have shown high efficiency of the proposed approach and superiority to the existing methods. [ABSTRACT FROM AUTHOR]

Stream-mining approach is defined as a set of cutting-edge techniques designed to process streams of data in real time, in order to extract knowledge. In the particular case of classification, stream-mining has to adapt its behavior to the volatile underlying data distributions, what has been called concept drift. It is important to note that concept drift may lead to situations where predictive models become invalid and have therefore to be updated to represent the actual concepts that data poses. In this context, there is a specific type of concept drift, known as recurrent concept drift, where the concepts represented by data have already appeared in the past. In those cases the learning process could be saved or at least minimized by applying a previously trained model. To deal with the aforementioned scenario, meta-models can be used in the process of enhancing the drift detection mechanisms used by data stream algorithms, by representing and predicting when the change will occur. There are some real-world situations where a concept reappears, as in the case of intrusion detection systems (IDS), where the same incidents or an adaptation of them usually reappear over time. In these environments the early prediction of drift by means of a better knowledge of past models can help to anticipate to the change, thus improving efficiency of the model regarding the training instances needed. Furthermore, as a complement of meta-models, a mechanism to assess the similarity between classification models is also needed when dealing with recurrent concepts. In this context, when reusing a previously trained model a rough comparison between concepts is usually made, applying boolean logic. The introduction of fuzzy logic comparisons between models could lead to a better efficient reuse of previously seen concepts, by applying not just equal models, but also similar ones. This work faces the aforementioned open issues by means of the MM-PRec system, that integrates a meta-model mechanism and a fuzzy similarity function. The theoretical proposal of MM-PRec is also validated in this paper by means of different experiments using both synthetic and real datasets. [ABSTRACT FROM AUTHOR]

Clustering used to discover hidden patterns in unlabeled data sets is an important task in data mining. Therefore, clustering validation applied to evaluating the clustering results has been recognized as one of the vital issues for clustering applications. The existing validation index is often used to evaluate the results rather than guide the clustering process dynamically. However, using the index can automatically adjust the algorithm's operation according to the actual data distribution, thereby improving the algorithm's adaptability. Simulating the division and aggregation process of cells in biology, in this paper, we propose a new two-phase (grouping and merging) cell group clustering algorithm by using a continuous validation and correction mechanism. In the grouping phase, a new clustering internal validation index called Split Index (S I) is utilized to evaluate the cohesion of a cell group continuously, and then a validation and correction mechanism is adopted to validate and split the cell groups so that the S I of each cell group can meet the split threshold ε , finally the cell nucleus of each cell group is determined by finding the sum of the minimum distances from the nucleus to other samples. In the merging phase, the cell group merge method is adopted to merge all the reachable cell groups in a density-reachable manner. Ultimately the clustering problem of arbitrarily distributed samples is completed. Experiments on the synthetic and the UCI Machine Learning Repository 1 1 http://archive.ics.uci.edu/ml/index.php. data set show that the validation index can effectively guide the clustering process, and the algorithm can deal with various data sets, including imbalanced data sets and spherical non-spherical clusters. [Display omitted] • We define a Split Index to evaluate the cohesion of a cell group. • A new method using clustering validation index is applied in clustering process. • Algorithm SICG is proposed based on validation & correction mechanism. • SICG with good result has little dependence on the parameter. [ABSTRACT FROM AUTHOR]

In the past two decades, the dimensionality of datasets involved in machine learning and data mining applications has increased explosively. Therefore, feature selection has become a necessary step to make the analysis more manageable and to extract useful knowledge about a given domain. A large variety of feature selection techniques are available in literature, and their comparative analysis is a very difficult task. So far, few studies have investigated, from a theoretical and/or experimental point of view, the degree of similarity/dissimilarity among the available techniques, namely the extent to which they tend to produce similar results within specific application contexts. This kind of similarity analysis is of crucial importance when two or more methods are combined in an ensemble fashion: indeed the ensemble paradigm is beneficial only if the involved methods are capable of giving different and complementary representations of the considered domain. This paper gives a contribution in this direction by proposing an empirical approach to evaluate the degree of consistency among the outputs of different selection algorithms in the context of high dimensional classification tasks. Leveraging on a proper similarity index, we systematically compared the feature subsets selected by eight popular selection methods, representatives of different selection approaches, and derived a similarity trend for feature subsets of increasing size. Through an extensive experimentation involving sixteen datasets from three challenging domains (Internet advertisements, text categorization and micro-array data classification), we obtained useful insight into the pattern of agreement of the considered methods. In particular, our results revealed how multivariate selection approaches systematically produce feature subsets that overlap to a small extent with those selected by the other methods. [ABSTRACT FROM AUTHOR]

• The collaboration between services were considered in recommendation. • An evaluation indicator named manufacturing service collaboration frequency indicator (CFI) were defined. • The CFI of service combinations were calculated by FP-growth and Simrank. • Optimizing service combinations with a MOEA combining CFI filtering (MOEA + CFIF). • MOEA + CFIF obtained solutions with both high CFI and QoS. The popularization of the service-oriented manufacturing mode makes increasing customers configure the required manufacturing services from Industry Internet platforms. However, most recommendation methods mainly focus on the QoS indicators of the overall manufacturing service combination without considering the collaboration relationship between manufacturing services. A novel multi-attribute recommendation method of manufacturing service combination considering the historical collaboration relationship of the online platforms is studied in this paper. First, an evaluation indicator named manufacturing service collaboration frequency indicator (CFI) is introduced, which uses an improved machine learning algorithm combining FP-growth and Simrank to mine the frequent terms and similarities of the manufacturing service collaboration process. Then, a multi-objective evolutionary algorithm considering CFI and QoS is proposed to recommend a more reliable combination of manufacturing services for customers. Finally, comparative experiments are conducted to demonstrate the effectiveness and practicability of our method. [ABSTRACT FROM AUTHOR]

Clustering has been applied in machine learning, data mining and so on, and has received extensive attention. However, since some data has noise or outliers, these noise or outliers easily bring about the objective function with large errors. In this paper, a robust clustering model with ℓ 2 , 1 , ℓ 1 norm and Laplacian regularization (RCLR) is proposed, on which, sparse error matrix is introduced to express sparse noise, and ℓ 1 norm is introduced to alleviate the sparse noise. In addition, the ℓ 2 , 1 norm is also introduced to achieves space robust by virtue of its nice rotation invariance property. Therefore, our RCLR is insensitive to data noise and outliers. More importantly, the Laplacian regularization is introduced into the RCLR to improve the clustering accuracy. In order to solve the optimization objective of clustering problem, we propose an iterative updating algorithm, named alternating direction method of multipliers (ADMM), to update each optimization variable alternatively, and the convergence of the proposed algorithm is also proved in theory. Finally, experimental results on a total of eleven datasets of three types of datasets, elaborate the superiority of this method over six existing classical clustering methods. Three types of datasets include face images dataset, handwritten recognition dataset, and UCI dataset. In particular, our RCLR clustering approach has the best effect on face image dataset. • Propose a joint optimization framework for clustering. • Utilize ℓ 2 , 1 and ℓ 1 to alleviate the influence of outliers and noises. • Prove algorithm convergence from theoretical and practical aspects. • Experimentation on three different types of real datasets. [ABSTRACT FROM AUTHOR]

Feature selection in real-world data mining problems is essential to make the learning task efficient and more accurate. Identifying the best feature selection algorithm, among the many available, is a complex activity that still relies heavily on human experts or some random trial-and-error procedure. Thus, the automated machine learning community has taken some steps towards the automation of this process. In this paper, we address the metalearning challenge of recommending feature selection algorithms by proposing a novel meta-feature engineering model. Our model considers a broad collection of meta-features that enable the study of the relationship between the dataset properties and the feature selection algorithm performance in terms of several criteria. We arrange the input meta-features into eight categories: (i) simple, (ii) statistical, (iii) information-theoretical, (iv) complexity, (v) landmarking, (vi) based on symbolic models, (vii) based on images, and (viii) based on complex networks (graphs). The target meta-features emerge from a multi-criteria performance measure, based on five individual performance indexes, that assesses feature selection methods grounded in information, distance, dependence, consistency, and precision measures. We evaluate our proposal using a recently developed framework that extracts the input meta-features from 213 benchmark datasets, and ranks the assessed feature selection algorithms, to fill in the target meta-features in meta-bases. This evaluation uses five state-of-the-art classification methods to induce recommendation models from meta-bases: C4.5, Random Forest, XGBoost, ANN, and SVM. The results showed that it is possible to reach an average accuracy of up to 90% applying our meta-feature engineering model. This work is the first to use an extensive empirical evaluation to provide a careful discussion of the strengths and limitations of more than 160 meta-features. These meta-features, while designed to aid the task of feature selection algorithm recommendation, can readily be employed in other metalearning scenarios. Therefore, we believe our findings are a valuable contribution to the fields of automated machine learning and data mining, as well as to the feature extraction and pattern recognition communities. • A novel meta-feature engineering model recommends feature selection algorithms. • The proposal obtains promising results from 213 datasets with hit rates of up to 90%. • Some simple, landmarking, image, and graph-based input meta-features highlighted. • A multi-criteria performance measure rigorously assesses candidate algorithms. • Chains of binary or multiclass classifiers can efficiently rank candidate algorithms. [ABSTRACT FROM AUTHOR]

• Discretization is an important process in terms of both machine learning and data mining. • The effect of discretization on ensemble methods was analyzed. • Four Chi-square based discretization algorithms were used. • The Ensemble methods performed better on discrete data sets. Classification is one of the important tasks in data mining and machine learning. Classification performance depends on many factors as well as data characteristics. Some algorithms are known to work better with discrete data. In contrast, most real-world data contain continuous variables. For algorithms working with discrete data, these continuous variables must be converted to discrete ones. In this process called discretization, continuous variables are converted to their corresponding discrete variables. In this paper, four Chi-square based supervised discretization algorithms ChiMerge(ChiM), Chi2, Extended Chi2(ExtChi2) and Modified Chi2(ModChi2) were used. In the literature, the performance of these algorithms is often tested with decision trees and Naïve Bayes classifiers. In this study, differently, four sets of data discretized by these algorithms were classified with ensemble methods. Classification accuracies for these data sets were obtained through using a stratified 10-fold cross-validation method. The classification performance of the original and discrete data sets of the methods is presented comparatively. According to the results, the performance of the discrete data is more successful than the original data. [ABSTRACT FROM AUTHOR]