Search

Finding dependencies in the data requires the analysis of relations between dozens of parameters of the studied process and hundreds of possible sources of influence on this process. Dependencies are nondeterministic and therefore modeling requires the use of statistical methods for analyzing random processes. Part of the information is often hidden from observation or not monitored. That is why many difficulties have arisen in the process of analyzing the collected information. The paper aims to find frequent patterns and parameters affected by COVID-19. The novelty of the paper is hierarchical architecture comprises supervised and unsupervised methods. It allows the development of an ensemble of the methods based on k-means clustering and classification. The best classifiers from the ensemble are random forest with 500 trees and XGBoost. Classification for separated clusters gives us higher accuracy on 4% in comparison with dataset analysis. The proposed approach can be used also for personalized medicine decision support in other domains. The features selection allows us to analyze the following features with the highest impact on COVID-19: age, sex, blood group, had influenza.

In recent years, more and more anonymous network have been developed. Since user's identity is difficult to trace in anonymous networks, many illegal activities are carried out in darknet. In this paper, we propose a hierarchical classifier of darknet traffic which can distinguish four types of darknet(Tor, I2P, ZeroNet, Freenet) and 25 darknet users' behavior. Due to the lack of public datasets, we deployed a darknet data probe that can capture real darknet traffic in Tor, I2P, ZeroNet, Freenet. After collecting and labeling darknet traffic, we extract 26 time-based flow features that can represent the characteristics of darknet traffic and train a hierarchical classifier constructed by 6 local classifiers. Results show that the classifier can easily distinguish Tor, I2P, ZeroNet, Freenet four kinds of darknet clients with an accuracy of 96.9% and identify 8 kinds of user behaviors for each type of darknet with an accuracy of 91.6% on average. With the help of this hierarchical classification method, darknet user behaviors can be accurately distinguished at the traffic exit.

Corporate companies are becoming more aware of gathering public sentiment, which is facilitated with the presence and vast usage of social networks and media platforms. This is a big data problem, and thus automated machine learning systems are deployed. The process requires the analysis of textual mentions, visual illustrations of the brand and/or respective location, as well as audio mentions of the corporate identity and respective products. When focusing on gathering sentiment analysis from the spoken language, the problem of accent recognition is evident across native and non-native English speakers. Thus, in this research, we investigate the key features of accent recognition, calibrate a proposed system based on previous research using the Wildcat Corpus, and apply on a recent dataset, the Common Voice. Finally applying to a custom dataset gathered from an online media platform. We propose a novel hierarchical classifier solution, trained on the Common Voice dataset and tested on the custom dataset. Our three-tier solution achieved 86% and 89% in the first two levels of accents, and 59% at the final level. From this research, we highlight the issues around the considered datasets and propose a number of recommendations for future researchers. In this research we are not improving or comparing any existing works, but rather offer new insights on the Common Voice dataset. We are presenting a hierarchical classifier for the accent classification problem as proposed.

Taxonomy of living organisms gains major importance in making the study of vastly heterogeneous living things easier. In addition, various fields of applied biology (e.g., agriculture) depend on classification of living creatures. Specific fragments of the DNA sequence of a living organism have been defined as DNA barcodes and can be used as markers to identify species efficiently and effectively. The existing DNA barcode-based classification approaches suffer from three major issues: (i) most of them assume that the classification is done within a given taxonomic class and/or input sequences are prealigned, (ii) highly performing classifiers, such as SVM, cannot scale to large taxonomies due to high memory requirements, (iii) mutations and noise in input DNA sequences greatly reduce the taxonomic classification accuracy. In order to address these issues, we propose a multi-level hierarchical classifier framework to automatically assign taxonomy labels to DNA sequences. We utilize an alignment-free approach called spectrum kernel method for feature extraction. We build a proof-of-concept hierarchical classifier with two levels, and evaluated it on real DNA sequence data from BOLD systems. We demonstrate that the proposed framework provides higher accuracy than regular classifiers. Besides, hierarchical framework scales better to large datasets enabling researchers to employ classifiers with high accuracy and high memory requirement on large datasets. Furthermore, we show that the proposed framework is more robust to mutations and noise in sequence data than the non-hierarchical classifiers.

Memotion analysis is a very crucial and important subject in today’s world that is dominated by social media. This paper presents the results and analysis of the SemEval-2020 Task-8: Memotion analysis by team Kraken that qualified as winners for the task. This involved performing multimodal sentiment analysis on memes commonly posted over social media. The task comprised of 3 subtasks, Task A was to find the overall sentiment of a meme and classify it into positive, negative or neutral, Task B was to classify it into the different types which were namely humour, sarcasm, offensive or motivation where a meme could have more than one category, Task C was to further quantify the classifications achieved in task B. An imbalanced data of 6992 rows was utilized for this which contained images (memes), text (extracted OCR) and their annotations in 17 classes provided by the task organisers. In this paper, the authors proposed a hybrid neural Naive-Bayes Support Vector Machine and logistic regression to solve a multilevel 17 class classification problem. It achieved the best result in Task B i.e 0.70 F1 score. The authors were ranked third in Task B.

Self-driving cars and Advanced Driver Assistance Systems rely heavily on Traffic Sign Recognition for safe maneuvering on the roads. But traffic signs can vary from one country to another, thereby necessitating multiple classifiers or a single universal classifier which can handle variations across countries. This paper reports our attempt at building a universal classifier. This classifier has to deal with large intra-class variations in the classes and also similarities among various difficult to distinguish traffic sign classes. This paper is an extension of our previous work in which we proposed a hierarchical classifier for traffic signs of a specific country. In hierarchical classification, dedicated classifiers are trained for classes which are more difficult to distinguish. Such similar classes are grouped together automatically by learning category hierarchy from the confusion matrix of a flat classifier (building block). In this paper, we use attention network for country independent classification. Here, CNN itself pays attention to regions in an image which are more discriminative and thus results in better classification for such problems. The aim here is to design a traffic sign recognition framework which can be used for multiple countries and be able to classify even the hard to distinguish classes by exploiting category hierarchy of traffic signs. The model is evaluated on traffic signs of seven countries namely Belgium, China, Croatia, Russia, Spain, Germany and Italy. The new building block architecture shows significant improvement of classification accuracy that is 97.7% as compared to building block architecture (VGG) used in our previous paper that is 95.1%.

Categories of Point of Interest (POI) facilitate location-based services from many aspects like location search and place recommendation [6]. However, POI categories are often incomplete and new POIs are increasing, this rises the problem of automatic POI classification. Current POI classification methods suffer from two problems: lack of textual information about POIs and not leveraging the hierarchical structure of the categories. In this paper, we propose an Ensemble POI Hierarchical Classification framework (EHC) consisting of three components: Textual and Geographic Feature Extraction, Hierarchical Classifier, and Soft Voting Ensemble Model. We conduct extensive experiments to demonstrate the effectiveness of our framework.

The availability of remote sensing images of various resolutions has enabled the incorporation of landscape structures in land-cover mapping. Despite the effectiveness of landscape metrics in quantifying landscape structures, they are inadequate in characterizing three elements: spatial neighborhoods, spatial dependencies, and semantic dependencies. Moreover, methods for mining the regularity of landscape-structure heterogeneity (i.e., spatial variations in landscape structures) are still limited, particularly for applications in urban land-cover mapping. This study hence proposes a novel approach with the aims to (1) characterize landscape structures considering the above three elements; (2) mine the regularity of landscape-structure heterogeneity; and (3) apply landscape-structure information as contexts to improve urban land-cover mapping. To achieve the first aim, landscape-structure features including pair-wise spatial relationships and neighborhood-based landscape metrics are defined. To accomplish the second aim, a clustering technique and a landscape infographic are used to cluster landscape structures and visualize landscape-structure types, respectively. Finally, a hierarchical classifier based on the feedforward multi-layer perceptron is developed for the third aim. Experiments are conducted in a heterogeneous urban environment in Beijing, China. The results show that the proposed approach, which considers 34 landscape-structure features and 19 landscape-structure types, achieves a classification accuracy improvement of 6.43% compared with the approaches without considering landscape-structure information. This study therefore demonstrates the effectiveness of incorporating landscape-structure features and landscape-structure types in improving urban land-cover mapping.

With smartphones increasingly becoming ubiquitous and being equipped with various sensors, nowadays, there is a trend towards implementing HAR (Human Activity Recognition) algorithms and applications on smartphones, including health monitoring, self-managing system and fitness tracking. However, one of the main issues of the existing HAR schemes is that the classification accuracy is relatively low, and in order to improve the accuracy, high computation overhead is needed. In this paper, an efficient Group-based Context-aware classification method for human activity recognition on smartphones, GCHAR is proposed, which exploits hierarchical group-based scheme to improve the classification efficiency, and reduces the classification error through context awareness rather than the intensive computation. Specifically, GCHAR designs the two-level hierarchical classification structure, i.e., inter-group and inner-group, and utilizes the previous state and transition logic (so-called context awareness) to detect the transitions among activity groups. In comparison with other popular classifiers such as RandomTree, Bagging, J48, BayesNet, KNN and Decision Table, thorough experiments on the realistic dataset (UCI HAR repository) demonstrate that GCHAR achieves the best classification accuracy, reaching 94.1636%, and time consumption in training stage of GCHAR is four times shorter than the simple Decision Table and is decreased by 72.21% in classification stage in comparison with BayesNet.

Age estimation from face images is a complex process as it varies from person to person, affected by various intrinsic factors (such as genetic and hormonal) and extrinsic factors (such as environmental, lifestyle, illumination, pose and expression). In this paper, an age estimation system has been proposed that preserves personalized aging trait as well as being robust to change in appearance, shape, wrinkle, texture, expression, pose and illumination of a face. The Golden ratio-based face cropping maintains uniformity of facial regions among faces irrespective of age, gender and race. Local, global and combinational features are extracted to handle the variations of intrinsic and extrinsic factors of aging. The experiment is conducted on FG-NET-AD and MORPH Album-2 facial aging databases. As the existing facial aging databases are incomplete, dealing with strong person-specificity, and high within-range variance; so the dominance of one age-group on other is resolved by feature filling that is carried out by a multi-feature-based modified expectation maximization algorithm. The age estimation is carried out by a three-level hierarchical classifier based on SVM and SVR by choosing suitable combination of hybrid feature sets. The experimental results show the superiority of proposed approach as compared to some of the existing age estimation approaches available in the literature.

In this paper, we propose hybrid principal component analysis (HPCA) to extract appearance feature of a face and inter-age group variation-based classifier (IAGVC) with regression to estimate age of a person. The proposed age estimation system is robust and less sensitive to outliers where nonuniform distribution of images at different age groups is existing. Under HPCA, we introduce two novel methods, extended SpPCA and extended SubXPCA. The issues, such as summarization of variance, variable component selection, computational complexity and classification accuracy of HPCA, have been addressed as well. The proposed HPCA operates on subpattern and whole pattern at a time and extracts appearance feature based on both local and global variation of faces. The IAGVC uses HPCA-based hybrid Eigen spaces of each training age group to estimate age group of a test image, and subsequently, support vector regressor estimates the specific age in the selective age group. The experimental results on FG-NET aging database show that the proposed HPCA-based IAGVC has better classification accuracy as compared to existing classical PCA, local SpPCA and SubXPCA over all age groups.

Daily human activity recognition using sensor data can be a fundamental task for many real-world applications, such as home monitoring and assisted living. One of the challenges in human activity recognition is to distinguish activities that have infrequent occurrence and less distinctive patterns. We propose a dissimilarity representation-based hierarchical classifier to perform two-phase learning. In the first phase, the classifier learns general features to recognise majority classes, and the second phase is to collect minority and subtle classes to identify fine difference between them. We compare our approach with a collection of state-of-the-art classification techniques on a real-world third-party dataset that is collected in a two-user home setting. Our results demonstrate that our hierarchical classifier approach outperforms the existing techniques in distinguishing users in performing the same type of activities. The key novelty of our approach is the exploration of dissimilarity representations and hierarchical classifiers, which allows us to highlight the difference between activities with subtle difference, and thus allows the identification of well-discriminating features. Postprint

We propose two optimization techniques to minimize memory usage and computation while meeting system timing constraints for real-time classification in wearable systems. Our method derives a hierarchical classifier structure for Support Vector Machine (SVM) in order to reduce the amount of computations, based on the probability distribution of output classes occurrences. Also, we propose a memory optimization technique based on SVM parameters, which results in storing fewer support vectors and as a result requiring less memory. To demonstrate the efficiency of our proposed techniques, we performed an activity recognition experiment and were able to save up to 35% and 56% in memory storage when classifying 14 and 6 different activities, respectively. In addition, we demonstrated that there is a trade-off between accuracy of classification and memory savings, which can be controlled based on application requirements.

Stance detection has gained increasing interest from the research community due to its importance for fake news detection. The goal of stance detection is to categorize an overall position of a subject towards an object into one of the four classes: agree, disagree, discuss, and unrelated. One of the major problems faced by current machine learning models used for stance detection is caused by a severe class imbalance among these classes. Hence, most models fail to correctly classify instances that fall into minority classes. In this paper, we address this problem by proposing a hierarchical representation of these classes, which combines the agree, disagree, and discuss classes under a new related class. Further, we propose a two-layer neural network that learns from this hierarchical representation and controls the error propagation between the two layers using the Maximum Mean Discrepancy regularizer. Compared with conventional four-way classifiers, this model has two advantages: (1) the hierarchical architecture mitigates the class imbalance problem; (2) the regularization makes the model to better discern between the related and unrelated stances. An extensive experimentation demonstrates state-of-the-art accuracy performance of the proposed model for stance detection.

Personalization related research has been conducted for decades with the objective of investigating personal-level effectiveness in irregular or abnormal cases among many types of domains. However, the activity recognition community generally investigates one-fits-all model, i.e., a single recognition model for all people and suffers from poor performance generalization or a huge amount of training data for high generalization. In this paper, we propose Compatibility-based classifier personalization (CbCP) as a subject-dependent activity recognition method that uses the classifier with the highest compatibility (similarity) with a particular user. Furthermore, we explore the effectiveness of classifier personalization through a particular group of activities, which results in a hierarchical recognition model. We present a comparative evaluation of 1) the traditional one-fits-all classifier vs. CbCP models and 2) compatibility evaluation through all activities (non-hierarchical classification) vs. a group of activities (hierarchical classification). The results of the two public datasets imply the effectiveness of compatibility-based approach for a hierarchical classifier formation.

A cancer of unknown primary (CUP) is a metastatic cancer for which standard diagnostic tests fail to locate the primary cancer. As standard treatments are based on the cancer type, such cases are hard to treat and have very poor prognosis. Using molecular data from the metastatic cancer to predict the primary site can make treatment choice easier and enable targeted therapy. In this article, we first examine the ability to predict cancer type using different types of omics data. Methylation data lead to slightly better prediction than gene expression and both these are superior to classification using somatic mutations. After using 3 data types independently, we notice some differences between the classes that tend to be misclassified, suggesting that integrating the data might improve accuracy. In light of the different levels of information provided by different omics types and to be able to handle missing data, we perform multi-omics classification by hierarchically combining the classifiers. The proposed hierarchical method first classifies based on the most informative type of omics data and then uses the other types of omics data to classify samples that did not get a high confidence classification in the first step. The resulting hierarchical classifier has higher accuracy than any of the single omics classifiers and thus proves that the combination of different data types is beneficial. Our results show that using multi-omics data can improve the classification of cancer types. We confirm this by testing our method on metastatic cancers from the MET500 dataset.

In data stream mining, state-of-the-art machine learning algorithms for the classification task associate each event with a class belonging to a finite, devoid of structural dependencies and usually small, set of classes. However, there are more complex dynamic problems where the classes we want to predict make up a hierarchal structure. In this paper, we propose an incremental method for hierarchical classification of data streams. We experimentally show that our stream hierarchical classifier present advantages to the traditional online setting in three real-world problems related to entomology, ichthyology, and audio processing.

Recognition of oil well drilling activities is a crucial task as it allows for identification of nonproductive time. Activity recognition is also important as it can be part of a complete oil well monitoring system. In this paper, we propose a novel hierarchical classifier that consists of Fuzzy Rule-based (FRB) and Random Forest (RF) Classifiers to recognize distinct oil well drilling activities in an oil well drilling process. The novel hierarchical classifier is designed by stacking the FRB and RF classifiers to achieve a classification of the numerous oil well drilling activities with high accuracy. The evaluation of the proposed method is illustrated through a case study of identifying the drilling activities in a real drilling data from oil well rig. Further recommended classifier generates an accurate report of time spent to execute the numerous drilling activities in a complete cycle of the oil well drilling process. Empirical evaluation of the real drilling data shows the efficiency and stability of the proposed method.

In this paper, we propose an algorithm for human activity recognition based on Gaussian Process Classifier (GPC). A hierarchical strategy is firstly applied to classify dynamic and static behaviors. Then, in each layer, three kinds of classification approaches are validated and evaluated for promoting recognition accuracy. Moreover, discriminative analysis method is invoked to cast high dimension features into lower dimensional space where classes are easily separated. Extensive experiments have been conducted and three vital points are observed: Firstly, GPC achieves comparable classification accuracy with other classifiers under the same experimental condition. Secondly, in case of less training samples, GPC outperforms the prominent Support Vector Machine (SVM) classifier. Thirdly, unlike SVM, GPC is more robust to the high dimensional features. Furthermore, we successfully implement the presented recognition algorithm into our hardware platform and achieve 99.75% accuracy on average in dealing with four sample activities.

In energy applications, fuels are processed in various ways according to their types. Experiments conducted with non-optimal processing procedures cause waste of the materials and may lead to inaccurate conclusions. Moreover, classifying the fuel material becomes harder when the material is already processed or collected from an environment that makes it difficult to discern. Thus, with the constantly developing and diversifying energy applications, the need for a framework that can classify solid fuels is increased. In this study, we used machine learning based approach to classify fuels with the use of proximate analysis results, i.e., fixed carbon, volatile matter and ash contents. We collected a data set from the literature and group them into four classes, i.e., coals, woods, agricultural residue and manufactured biomass. Then, K-nearest neighbor, support vector machine and random forest machine learning classifiers are employed to develop prediction models. Furthermore, hierarchical classification approach is taken to combine each classifier’s advantages with integration of expert opinion to create a complete and highly accurate classifier framework which can classify an unknown fuel into one of the four categories. K-fold cross validation is used to evaluate classifiers’ performances in unbiased manner. With hierarchical classifier, % 96 and %92 classification accuracy is achieved for training and testing phases, respectively. Source code of the proposed hierarchical classifier framework is also provided.

The paper proposes a Machine Learning methodology for Android malware detection and recognition, including crypto-mining applications using the blockchain. The design is based on a hierarchical classification method, with several decision stages. A combination of functional and statistical features is proposed to be applied for data classification in order to provide a high-performance malware recognition process. The specific contribution of this design methodology is the hierarchical classifier with detection and discrimination stages, respectively. Further works should be done for various features sets in order to achieve an optimized and high-accuracy modeling process supporting an innovative Machine Learning-based solution for Android malware detection.

Authorship Attribution concerns with the classification problem of identifying the original author of a piece of text by analyzing other literary works of the same author. A substantial amount of approaches have been taken to solve this problem in Bengali Literature, including machine learning techniques and natural language processing methods. Most of the classification approaches, especially the machine learning methods, have a high computational requirement and time complexity for model training and testing whenever a large amount of data is to be processed. This restricts their uses in everyday electronic devices and computers with lower system specifications. In this paper, we take a different approach in exploring the classification problem using Facebook's open source fastText technology, specifically due to its low computational power requirement and faster model training, to check the possibility of its use in applications in regular computers and mobile systems. Our approach with fastText provided us with convincing results, outperforming traditional machine learning based classifiers, m ore specifically, Naive Bayes' accuracy whenever longer length n-grams were taken as feature sets. FastText also trained its classification models much faster than those of Support Vector Machines (SVM) and Naive Bayes.

Prognostics and health management (PHM) is the study of using health information to support decision making to improve maintenance and operations. There are many existing methods for PHM but most solely focus on predictive accuracy and ignore resource constraints. In a real-world application of a PHM system, resources consumed by the predictive algorithm at the core of the PHM system could be a limiting factor. In this study, we propose using a hierarchical classification scheme to break the conventional classification problem into many sub-problems arranged in a hierarchy. By splitting the diagnostic task into many sub-problems, the hierarchical classifier can be constructed to maximize accuracy while minimizing resource consumption. Reinforcement learning is proposed to select the classifiers for each sub-problem. The proposed methodology is applied to condition monitoring of a hydraulic actuator where power is a limiting resource. Numerical experiments demonstrate that the proposed hierarchical classification method can reduce resource consumption compared to a traditional flat classification approach.

Territorial planning and management requires that the spatial structure of the socioecological sectors is adequately understood. Several classification techniques exist that have been applied to detect ecological, or socioeconomic sectors, but not simultaneously in the same model; and also, with a limited number of variables. We have developed and applied a new probabilistic methodology – based on hierarchical hybrid Bayesian network classifiers – to identify the different socioecological sectors in Andalusia, a region in southern Spain, and incorporate a scenario of change. Results show that a priori , the socioecological structure is highly heterogeneous, with an altitude gradient from the river basin to the mountain peaks. However, under a scenario of global environmental change this heterogeneity is lost, making the territory more vulnerable to any alteration or disturbance. The methodology applied allows dealing with complex problems, containing a large number of variables, by splitting them into several sub-problems that can be easily solved. In the case of territorial planning, each component of the territory is modelled independently before combining them into a general classifier model. Furthermore, it can be applied to any complex unsupervised classification problem with no modification to the methodology.

An image-driven, model-free approach to design control systems for a large class of industrial process is proposed. A mathematical model of the process is replaced by sequences of subsequent images which play the role of the process (plant) states. The length of this sequences depends on the speed of the process dynamics and on the frame rate. Firstly, a learning sequence of the system states is collected and then, it is used for classifying (clustering) its states. A decision of the control system is attached by an expert to each class (cluster) of the states. At the implementation stage images from a camera in the loop are collected into subsequences corresponding to the system states, then they are classified and a control action corresponding to a class at hand is undertaken. This general idea is exemplified by the case study of a laser power control in an additive manufacturing, which is a repetitive process. A tree-like, hierarchical classifier is proposed in order to recognize the process states, each consisting of three consecutive images. Its performance is tested on real-life images from the process of laser cladding additive manufacturing.

Human activity recognition has been gaining more and more attention from researchers in recent years, particularly with the use of widespread and commercially available devices such as smartphones. However, most of the existing works focus on discriminative classifiers while neglecting the inherent time-series and continuous characteristics of sensor data. To address this, we propose a two-stage continuous hidden Markov model framework, which also takes advantage of the innate hierarchical structure of basic activities. This kind of system architecture not only enables the use of different feature subsets on different subclasses, which effectively reduces feature computation overhead, but also allows for varying number of states and iterations. Experiments show that the hierarchical structure dramatically increases classification performance. We analyze the behavior of the accelerometer and gyroscope signals for each activity through graphs, and with added fine tuning of states and training iterations, the proposed method is able to achieve an overall accuracy of up to 93.18%, which is the best performance among the state-of-the-art classifiers for the problem at hand.

We propose a passive multi-purpose scheme for photographic image (PIM) detection and a device class identification method. The motivation for the scheme is the periodicity phenomenon caused by color filter arrays (CFAs) and the demosaicing process. The phenomenon only occurs in the Fourier spectrum in PIMs. The proposed scheme exploits prediction error statistics, local peak detection, and a PIM classifier to analyze the phenomenon for PIM detection. We also develop a hierarchical classifier for device class identification based on the analysis of local peaks in the Fourier spectrum. To evaluate the scheme's performance, we compile a test dataset of PIMs and PRCG (photorealistic computer graphics) images, and analyze the impact of leak peak detection, JPEG lossy compression, and cropping operations on PIM detection. The accuracy rate of the scheme on 5805 test images is 95.56%, which is higher than that of the methods proposed in Sutthiwan et al. (2009) [1] and Gallagher and Chen (2008) [2]. In addition, for device class identification, the precision rate of the proposed method is at least 93% on Canon, Sony, and Nikon images. The experiment results demonstrate the efficacy of the proposed multi-purpose scheme.

In Jones and Goel (2012), we describe a meta-reasoning architecture that uses abstraction networks (ANs) and empirical verification procedures (EVPs) to ground self-diagnosis and self-repair of domain knowledge in perception. In particular, we showed that when a hierarchical classifier organized as an AN makes an incorrect prediction, then meta-reasoning can help diagnose and repair the semantics of the concepts in the network. Further, we demonstrated that if an EVP associated with each concept in the network can verify the semantics of that concept at diagnosis time, then the meta-reasoner can perform knowledge diagnosis and repair tractably. In this article, we report on three additional results on the use of perceptually grounded meta-reasoning for correcting prediction errors. Firstly, a new theoretical analysis indicates that the meta-reasoning diagnostic procedure is optimal and establishes the knowledge conditions under which the learning converges. Secondly, an empirical study indicates that the EVPs themselves can be adapted through refining the conceptual semantics. Thirdly, another empirical study shows that if EVPs cannot be defined for all concepts in a hierarchy, the computational technique degrades gracefully. While the theoretical analysis provides a deeper explanation of the sources of power in ANs, the two empirical studies demonstrate ways in which the strong assumptions made by ANs in their most basic form can be relaxed.

We propose a method to recognize pollen grains using a two-stage classifier. First, texture classification categorizes the pollen grains into sub-groups. Then, a final classification of individual pollen types is done by segmenting the image int multiple layers of regions for each pollen image. The main novelty in our method is threefold: (1) Adopting two successive classification stages. (2) Combining hierarchical clustering and SVM algorithms to merge similar pollen types into sub-groups. (3) Adopting a layering approach prior to performing feature extraction. The combination of these aspects gives excellent results. We evaluated our method using 1,063 light-microscopy images of pollen grains from 30 species. The results show that: (1) the layering technique increases the classification rate by almost almost 7% over using the same features directly. (2) adopting two classification stage increases the classification rate by 6%. (3) the proposed system outperformed traditional techniques.

This paper proposes a multi-level meta-classifier for identifying human activities based on accelerometer data. The training data consists of 77 subjects performing a combination of 23 different activities and monitored using a single hip-worn triaxial accelerometer. Time and frequency based features were extracted from two-second windows of raw accelerometer data and a subset of the features, together with demographic information, was selected for classification. The activities were divided into five activity groups: non-ambulatory activities, walking, running, climbing upstairs, and climbing downstairs. Multiple classification techniques were tested for each classifier level and groups. Random forests were found to perform comparatively better at each level. Based upon those tests, a 3-level hierarchical classifier, consisting of 5 random forest classifiers, was built. At the first level, the non-ambulatory activities are separated from the rest. At the second, the ambulatory activities are divided into four activity groups. At the final level, the activities are classified individually. Accuracy on test sets was found to be approximately 87% overall for individual activities and 94% at the activity group level. These results compare favorably to contemporary results in classifying human activity.

Fingerprint classification is still a challenging problem due to large intra-class variability, small inter-class variability and the presence of noise. To deal with these difficulties, we propose a regularized orientation diffusion model for fingerprint orientation extraction and a hierarchical classifier for fingerprint classification in this paper. The proposed classification algorithm is composed of five cascading stages. The first stage rapidly distinguishes a majority of Arch by using complex filter responses. The second stage distinguishes a majority of Whorl by using core points and ridge line flow classifier. In the third stage, K-NN classifier finds the top two categories by using orientation field and complex filter responses. In the fourth stage, ridge line flow classifier is used to distinguish Loop from other classes except Whorl. SVM is adopted to make the final classification in the last stage. The regularized orientation diffusion model has been evaluated on a web-based automated evaluation system FVC-onGoing, and a promising result is obtained. The classification method has been evaluated on the NIST SD 4. It achieved a classification accuracy of 95.9% for five-class classification and 97.2% for four-class classification without rejection.

The performance of an efficient and accurate face detection system depends on several issues: (1) distinctive representation for face patterns; (2) effective algorithm for feature selection and classifier learning; (3) suitable framework for rapid background removal. To address the first issue, we propose to represent face patterns with a set of heterogeneous and complementary feature descriptors including the Generalized Haar-like (GH) descriptor, Multi-Block Local Binary Patterns (MB-LBP) descriptor and Speeded-Up Robust Features (SURF) descriptor. To address the second issue, Particle Swarm Optimization (PSO) algorithm is incorporated into the Adaboost framework, replacing the exhaustive search used in original Adaboost for efficient feature selection. The utilization of heterogeneous feature descriptors enriches the diversity of feature types for Adaboost learning algorithm. As a result, classification performance of the boosted ensemble classifier also improves significantly. A three-stage hierarchical classifier structure is proposed to tackle the last issue. In particular, a new stage is added to detect candidate face regions more quickly by using a large size window with a large moving step. Nonlinear support vector machine (SVM) classifiers are used instead of decision stump classifiers in the last stage to remove those remaining complex non-face patterns that cannot be rejected in the previous two stages. Combining the abovementioned effective modules, we derive the proposed Hetero-PSO-Adaboost-SVM face detector that achieves superior detection accuracy while maintaining a low training and detection complexity. Extensive experiments demonstrate the robustness and efficiency of our system by comparing it with several popular state-of-the-art algorithms on our own test set as well as the widely used CMU+MIT frontal and CMU profile face dataset.

Predicting the future actions of individuals from geospatial data has the potential to provide a basis for tailored services. This work presents the Predictive Context Tree (PCT), a new hierarchical classifier based on the Context Tree summary model [8]. The PCT is capable of predicting the future contexts and locations of individuals to provide a basis for understanding not only where a user will be, but also what type of activity they will be performing. Through a comparison to established techniques, this paper demonstrates the applicability of the PCT by showing increased accuracies for location prediction, and increased utility through context prediction.

The flourishing fame and development of big data in recent years made researchers to have a detailed study. Of the all entire emerging big data research topics, classification of data from big data is identified as a great challenge to address as of our analysis. The Classification is the process of categorizing data for its most effective and efficient use. While analyzing large scale patient records, hierarchical learning approach which is tree structured that train max-margin classifier will give better classification results and also it is computationally efficient. The quality of features has an effect on the performance of hierarchical learning approach for classification of patient records. So we have to extract discriminative features for training hierarchical classifier. In this paper Highly Correlated Feature Set Selection (HCFS) algorithm is proposed to combine with the hierarchical leaning approach to improve its performance. This algorithm identifies the good feature subsets which will improve the classification accuracy.

The problem of automatic classification of seven types of eating conditions from speech is considered. Based on the confusion among different eating conditions from a seven class support vector machine (SVM) classifier, a hierarchical SVM classifier is designed. Experiments on the iHEARu-EAT database show that the hierarchical classifier results in a better classification accuracy compared to a seven class classifier. We also perform a feature selection for each of the classifiers in the hierarchical approach. This further improves the unweighted average recall (UAR) to 73.7% compared to an UAR of 60.9% obtained from the baseline scheme of a direct seven-way classification.

Social media are more and more frequently used by people to express their feelings in the form of short messages. This has raised interest in emotion detection, with a wide range of applications among which the assessment of users' moods in a community is perhaps the most relevant. This paper proposes a comparison between two approaches to emotion classification in tweets, taking into account six basic emotions. Additionally, it proposes a completely automated way of creating a reliable training set, usually a tedious task performed manually by humans. In this work, training datasets have been first collected from the web and then automatically filtered to exclude ambiguous cases, using an iterative procedure. Test datasets have been similarly collected from the web, but annotated manually. Two approaches have then been compared. The first is based on a direct application of a single "flat" seven-output classifier, which directly assigns one of the emotions to the input tweet, or classifies it as "objective", when it appears not to express any emotion. The other approach is based on a three-level hierarchy of four specialized classifiers, which reflect a priori relationships between the target emotions. In the first level, a binary classifier isolates subjective expressing emotions from objective tweets. In the second, another binary classifier labels subjective tweets as positive or negative. Finally, in the third, one ternary classifier labels positive tweets as expressing joy, love, or surprise, while another classifies negative tweets as expressing anger, fear, or sadness. Our tests show that the a priori domain knowledge embedded into the hierarchical classifier makes it significantly more accurate than the flat classifier.

The missing value imputation process can be defined as a preprocessing step that fills missing values of attributes in incomplete datasets. Nowadays, the problem of incomplete datasets in the hierarchical classification scenario must be solved using unsupervised missing value imputation methods due to the lack of supervised methods to deal with the hierarchical context. Thus, in this work, we propose and evaluate a supervised missing value imputation method for datasets used in hierarchical classification problems in which the classes are organized into tree structure. Experiments were performed on incomplete datasets to evaluate the effect of the proposed missing value imputation method on classification performance when using a global hierarchical classifier. The results showed that, using the proposed method for dealing with missing attribute values, it provided higher classifier predictive performance than other unsupervised missing value imputation methods.

This paper proposes a classification framework based on simple classifiers organized in a tree-like structure. It is observed that simple classifiers, even though they have high error rate, find similarities among classes in the problem domain. The authors propose to trade on this property by recognizing classes that are mistaken and constructing overlapping subproblems. The subproblems are then solved by other classifiers, which can be very simple, giving as a result a hierarchical classifier (HC). It is shown that HC, together with the proposed training algorithm and evaluation methods, performs well as a classification framework. It is also proven that such constructs give better accuracy than the root classifier it is built upon.

The analysis of daily living human behavior has proven to be of key importance to prevent unhealthy habits. The diversity of activities and the individuals' particular execution style determine that several sources of information are normally required. One of the main issues is to optimally combine them to guarantee performance, scalability and robustness. In this paper we present a fusion classification methodology which takes into account the potential of the individual decisions yielded at both activity and sensor classification levels. Particularly tested on a wearable sensors based system, the method reinforces the idea that some parts of the body (i.e., sensors) may be specially informative for the recognition of each particular activity, thus supporting the ranking of the decisions provided by each associated sensor decision entity. Our method systematically outperforms the results obtained by traditional multiclass models which otherwise may require a high-dimensional feature space to acquire a similar performance. The comparison with other activity-recognition fusion approaches also demonstrates our model scales significantly better for small sensor networks.

In most cases, the main goal of machine learning and data mining applications is to obtain good classifiers. However, final users, for instance researchers in other fields, sometimes prefer to infer new knowledge about their domain that may be useful to confirm or reject their hypotheses. This paper presents a learning method that works along these lines, in addition to reporting three interesting applications in the field of population genetics in which the aim is to discover relationships between species or breeds according to their genotypes. The proposed method has two steps: first it builds a hierarchical clustering of the set of classes and then a hierarchical classifier is learned. Both models can be analyzed by experts to extract useful information about their domain. In addition, we propose a new method for learning the hierarchical classifier. By means of a voting scheme employing pairwise binary models constrained by the hierarchical structure, the proposed classifier is computationally more efficient than previous approaches while improving on their performance.

This study comprehensively evaluates the performance of five types of probabilistic hierarchical classification methods used for predicting Gene Ontology (GO) terms related to ageing. Of those tested, a new hybrid of a Local Hierarchical Classifier (LHC) and the Predictive Clustering Tree algorithm (LHC-PCT) had the best predictive accuracy results. We also tested the impact of two types of variations in most hierarchical classification algorithms, namely: (a) changing the base algorithm (we tested Naive Bayes and Support Vector Machines), and the impact of (b) using or not the Correlation based Feature Selection (CFS) algorithm in a pre-processing step. In total, we evaluated the predictive performance of 17 variations of hierarchical classifiers across 15 datasets of ageing and longevity-related genes. We conclude that the LHC-PCT algorithm ranks better across several tests (seven out of 12). In addition, we interpreted the models generated by the PCT algorithm to show how hierarchical classification algorithms can be used to extract biological insights out of the ageing-related datasets that we compiled.

The research related to age estimation using face images has become increasingly important, due to the fact it has a variety of potentially useful applications. An age estimation system is generally composed of aging feature extraction and feature classification; both of which are important in order to improve the performance. For the aging feature extraction, the hybrid features, which are a combination of global and local features, have received a great deal of attention, because this method can compensate for defects found in individual global and local features. As for feature classification, the hierarchical classifier, which is composed of an age group classification (e.g. the class of less than 20 years old, the class of 20-39 years old, etc.) and a detailed age estimation (e.g. 17, 23 years old, etc.), provide a much better performance than other methods. However, both the hybrid features and hierarchical classifier methods have only been studied independently and no research combining them has yet been conducted in the previous works. Consequently, we propose a new age estimation method using a hierarchical classifier method based on both global and local facial features. Our research is novel in the following three ways, compared to the previous works. Firstly, age estimation accuracy is greatly improved through a combination of the proposed hybrid features and the hierarchical classifier. Secondly, new local feature extraction methods are proposed in order to improve the performance of the hybrid features. The wrinkle feature is extracted using a set of region specific Gabor filters, each of which is designed based on the regional direction of the wrinkles, and the skin feature is extracted using a local binary pattern (LBP), capable of extracting the detailed textures of skin. Thirdly, the improved hierarchical classifier is based on a support vector machine (SVM) and a support vector regression (SVR). To reduce the error propagation of the hierarchical classifier, each age group classifier is designed so that the age range to be estimated is overlapped by consideration of false acceptance error (FAE) and false rejection error (FRE) of each classifier. The experimental results showed that the performance of the proposed method was superior to that of the previous methods when using the BERC, PAL and FG-Net aging databases.

In hierarchical classification, classes are arranged in a hierarchy represented by a tree or a forest, and each example is labeled with a set of classes located on paths from roots to leaves or internal nodes. In other words, both multiple and partial paths are allowed. A straightforward approach to learn a hierarchical classifier, usually used as a baseline method, consists in learning one binary classifier for each node of the hierarchy; the hierarchical classifier is then obtained using a top-down evaluation procedure. The main drawback of this na¨ıve approach is that these binary classifiers are constructed independently, when it is clear that there are dependencies between them that are motivated by the hierarchy and the evaluation procedure employed. In this paper, we present a new decomposition method in which each node classifier is built taking into account other classifiers, its descendants, and the loss function used to measure the goodness of hierarchical classifiers. Following a bottom-up learning strategy, the idea is to optimize the loss function at every subtree assuming that all classifiers are known except the one at the root. Experimental results show that the proposed approach has accuracies comparable to state-of-the-art hierarchical algorithms and is better than the na¨ıve baseline method described above. Moreover, the benefits of our proposal include the possibility of parallel implementations, as well as the use of all available well-known techniques to tune binary classification SVMs

The complexity of classification tasks in computer vision applications has increased, as these systems have been adopted in more real-world applications. One of the hardest tasks to model by a computer system is related to achieving a human task based on subjective perception of the environment or goods. This paper presents a fuzzy classifier aimed to perform a subjective classification task, using the same criteria considered by a human inspector. The task consists on the description and classification of a cosmetic defect presented in ophthalmic lenses. The human inspector is modeled by a fuzzy hierarchical rule classifier (FHRC). The goal of the first stage is to obtain the description of the defect constituents. This information is then analyzed by the second stage to provide the final classification. The performance of the FHRC is 91.83% of correct classification in the single blob case and 81.48% in the multiblob. This performance can be considered acceptable, assuming that a human inspector has an estimated performance of 85%, with the advantages that the proposed system does not suffer from the disadvantages of a human inspector. The system also provides information of the defect constituents, which can be used to improve the fabrication process.

The paper considers the problem of classification error in multistage pattern recognition. This model of classification is based primarily on the Bayes rule and secondarily on the notion of fuzzy numbers. In adopting a probability-fuzzy model two concepts of hierarchical rules are proposed. In the first approach the local criterion that denote the probabilities of misclassification for particular nodes of a tree is considered. In the second approach the global optimal strategy that minimises the mean probability of misclassification on the whole multistage recognition process is considered. A probability of misclassifications is derived for a multiclass hierarchical classifier under the assumption that the features at different nodes of the tree are class-conditionally statistically independent, and we have fuzzy information on object features instead of exact information. Numerical example of this difference concludes the work.

Obtaining training data for land cover classification using remotely sensed data is time consuming and expensive especially for relatively inaccessible locations. Therefore, designing classifiers that use as few labeled data points as possible is highly desirable. Existing approaches typically make use of small-sample techniques and semisupervision to deal with the lack of labeled data. In this paper, we propose an active learning technique that efficiently updates existing classifiers by using fewer labeled data points than semisupervised methods. Further, unlike semisupervised methods, our proposed technique is well suited for learning or adapting classifiers when there is substantial change in the spectral signatures between labeled and unlabeled data. Thus, our active learning approach is also useful for classifying a series of spatially/temporally related images, wherein the spectral signatures vary across the images. Our interleaved semisupervised active learning method was tested on both single and spatially/temporally related hyperspectral data sets. We present empirical results that establish the superior performance of our proposed approach versus other active learning and semisupervised methods.

In this paper a novel complex classifier architecture is proposed. The architecture has a hierarchical tree-like structure with simple artificial neural networks (ANNs) at each node. The actual structure for a given problem is not preset but is built throughout training. The training algorithm's ability to build the tree-like structure is based on the assumption that when a weak classifier (i.e., one that classifies only slightly better than a random classifier) is trained and examples from any two output classes are frequently mismatched, then they must carry similar information and constitute a sub-problem. After each ANN has been trained its incorrect classifications are analyzed and new sub-problems are formed. Consequently, new ANNs are built for each of these sub-problems and form another layer of the hierarchical classifier. An important feature of the hierarchical classifier proposed in this work is that the problem partition forms overlapping sub-problems. Thus, the classification follows not just a single path from the root, but may fork enhancing the power of the classification. It is shown how to combine the results of these individual classifiers.

Motivation: G protein-coupled receptors (GPCRs) play an important role in many physiological systems by transducing an extracellular signal into an intracellular response. Over 50% of all marketed drugs are targeted towards a GPCR. There is considerable interest in developing an algorithm that could effectively predict the function of a GPCR from its primary sequence. Such an algorithm is useful not only in identifying novel GPCR sequences but in characterizing the interrelationships between known GPCRs.Results: An alignment-free approach to GPCR classification has been developed using techniques drawn from data mining and proteochemometrics. A dataset of over 8000 sequences was constructed to train the algorithm. This represents one of the largest GPCR datasets currently available. A predictive algorithm was developed based upon the simplest reasonable numerical representation of the protein's physicochemical properties. A selective top-down approach was developed, which used a hierarchical classifier to assign sequences to subdivisions within the GPCR hierarchy. The predictive performance of the algorithm was assessed against several standard data mining classifiers and further validated against Support Vector Machine-based GPCR prediction servers. The selective top-down approach achieves significantly higher accuracy than standard data mining methods in almost all cases.Contact: m.davies@mail.cryst.bbk.ac.uk

A novel pairwise decision tree (PDT) framework is proposed for hyperspectral classification, where no partitions and clustering are needed and the original C-class problem is divided into a set of two-class problems. The top of the tree includes all original classes. Each internal node consists of either a set of class pairs or a set of class pairs and a single class. The pairs are selected by the proposed sequential forward selection (SFS) or sequential backward selection (SBS) algorithms. The current node is divided into next-stage nodes by excluding either class of each selected pair. In the classification, an unlabelled pixel is recursively classified into the next node, by excluding the less similar class of each node pair until the classification result is obtained. Compared to the single-stage classifier approach, the pairwise classifier framework and the binary hierarchical classifier (BHC), experiments on an Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) data set for a nine-class problem demonstrated the effectiveness of the proposed framework.