Search

The high-dimensional nature of medical data frequently results in suboptimal performance of machine learning models. Applying feature selection before classification is necessary to improve the performance of classifiers. Although evolutionary-based wrapper feature selection methods are acknowledged for their superior performance in exploring optimal feature subsets, they have been demonstrated to carry the risk of overfitting and a potential loss of efficient search capability in the later stages of evolution. To address these issues, we propose a generalized wrapper feature selection method called Two Populations Based Feature Selection (TPBFS), which incorporates dual populations evolving in reverse directions to improve convergence speed. It introduces a probability-based crossover operation to mitigate overfitting and a record list to systematically track and replace optimal individuals, which helps to avoid getting stuck in local optima during later stages of evaluation. The experimental results demonstrate that TPBFS is effective in reducing the dimensionality of various medical datasets while guaranteeing the performance of classifiers. [ABSTRACT FROM AUTHOR]

Android malware has become an increasingly important threat to individuals, organizations, and society, posing significant risks to data security, privacy, and infrastructure. As malware evolves in sophistication and complexity, the detection and mitigation of these malicious software instances have become more challenging and time consuming since the required number of features to identify potential malware can be very high. To address this issue, we have developed an effective feature selection methodology for malware detection in Android. The critical concern in the field of malware detection is the complexity of algorithms and the use of features that are used to detect malware. The present paper delivers a methodology for pre-processing datasets to select the most optimal features that will allow detecting malware, while maintaining very high accuracy. The proposed methodology has been tested on two real world datasets and the results indicate that the number of features is significantly reduced from 489 to between 19 and 28 for the first dataset and from 9503 to between 9 and 27 for the second dataset, whilst the accuracy is maintained as if all features were used. [ABSTRACT FROM AUTHOR]

Credit scoring models are crucial tools for lenders to assess credit risks. Researchers from academia and the financial industry have shown intense interest in these models. However, real credit datasets often have high dimensionality and class imbalance, making it challenging to develop accurate and effective credit scoring models. To address these challenges, a new approach called the Multiple-Optimized Ensemble Learning (MOEL) method has been proposed. In MOEL, a technique called Multiple Diverse Optimized Subsets (MDOS) generates multiple diverse optimized subsets from various weighted random forests. From each subset, more effective and relevant features are selected. Then, a new evaluation measure is applied to each subset to determine the more optimized subsets. These subsets are applied to a novel Mahalanobis-based oversampling (MOS) technique to provide balanced subsets for the base classifier, which lessens the detrimental effects of imbalanced datasets. Finally, a stacking-based ensemble method is applied to the balanced subsets for integration of the base models. The proposed model was evaluated against six high-dimensional imbalanced credit scoring datasets, and it outperformed state-of-the-art methods, exhibiting a mean rank of 1.5 and 1.333 in terms of F1_score and G-mean, respectively. [ABSTRACT FROM AUTHOR]

Feature subset selection problem is an NP hard problem and there is a need for computationally efficient algorithms that find near optimal feature subsets which improve the performance of a classifier. Two major challenges for feature subset selection are high-dimensional data, that is, data with a large number of features and large datasets. Scalability of the feature selection algorithms in terms of accuracy for high dimensional data and the time taken for large datasets are important issues. We propose a consensus clustering based approach to feature selection that addresses these issues. There exist many greedy feature ranking algorithms in the literature that are computationally efficient. Each algorithm assigns a different ranking order to the features. A consensus among these rankings may provide a feature ranking that performs well with respect to time as well as accuracy. The goal of this work is to propose efficient algorithms that work on small as well as large datasets. The contributions of this work include: i. A fast and scalable approach for feature selection Feature ranking based on consensus clustering(FRCC), has been designed using the available feature ranking algorithms from the literature. ii. A parallelizable version of FRCC, namely, Hybrid Feature Selection(HFS), is proposed to address the feature reduction in large datasets. The implementation results show that FRCC clearly outperforms many recent algorithms in the literature on small as well as large dimensional data sets. HFS has been implemented on datasets with lakhs of instances and dimensionality in hundreds and thousands. HFS proves to be very effective in terms of feature reduction and accuracy in comparison to the results obtained by recent algorithms in the literature. [ABSTRACT FROM AUTHOR]

Intrusion Detection System (IDS) is designed being help and safeguard IoT networks from potential threats. Distributed Denial of Service (DDoS) assaults are a pernicious kind of cyber-attacks causing server disruptions in modern cyber-security world. Detecting unauthorized and suspicious activities by observing data traffic flows is crucial for enhanced network security. So, this research paper proposes an innovative solution to IoT environment by designing effective intrusion detection module. It benefits from the working principle of different modules that are operated for data dimensionality reduction, feature optimization and deep classification to maximize network security by identifying normal and malicious traffic flows. The detection process commences with data pre-processing steps such as null set removal and redundant feature elimination that provide a clear and concise representation of the data. Next, we employ the Random Subset Feature Selection (RSFS) technique to minimize dimension of preprocessed information by eliminating duplicate or redundant features. The selected feature subsets are then used as the initial search space for the Mutation Boosted Golden Jackal Optimization (MBGJO) algorithm. It helps to predict optimal attributes that contribute most effectively to detection of different attack classes. Finally, the Light Gradient Boosting Machine (LGBM) algorithm is used to train the ideal feature set and detect various attack classes in the CIC-DDoS2019 dataset. By employing this algorithm, we ensure that our detection system remains scalable and capable of handling diverse attack scenarios. Experimental findings demonstrate that our IDS attains an impressive accuracy of 99.7%. Moreover, it surpasses other state-of-the-art mechanisms with regard to scalability and security. Our intrusion detection system thus provides an effective solution for safeguarding IoT networks against potential threats. [ABSTRACT FROM AUTHOR]

The coresets approach, also called subsampling or subset selection, aims to select a subsample as a surrogate for the observed sample and has found extensive application in large-scale data analysis. Existing coresets methods construct the subsample using a subset of rows from the predictor matrix. Such methods can be significantly inefficient when the predictor matrix is sparse or numerically sparse. To overcome this limitation, we develop a novel element-wise subset selection approach, called core-elements, for large-scale least squares estimation. We provide a deterministic algorithm to construct the core-elements estimator, only requiring an O (nnz (X) + r p 2) computational cost, where X is an n × p predictor matrix, r is the number of elements selected from each column of X, and nnz (·) denotes the number of non-zero elements. Theoretically, we show that the proposed estimator is unbiased and approximately minimizes an upper bound of the estimation variance. We also provide an approximation guarantee by deriving a coresets-like finite sample bound for the proposed estimator. To handle potential outliers in the data, we further combine core-elements with the median-of-means procedure, resulting in an efficient and robust estimator with theoretical consistency guarantees. Numerical studies on various synthetic and real-world datasets demonstrate the proposed method’s superior performance compared to mainstream competitors. [ABSTRACT FROM AUTHOR]

Background: Total neoadjuvant therapy (TNT) has been recommended by the National Comprehensive Cancer Network for treating locally advanced rectal cancer (LARC), but extremely rare studies have focused on establishing nomograms to predict the prognosis in these patients after TNT. We aimed to develop a nomogram to predict overall survival (OS) in rectal cancer patients who underwent TNT. Methods: In retrospective cohort study, we extract the data of the rectal cancer patients from the SEER database between 2010 and 2015, including demographic information and tumor characteristics. The cohort was divided into training set and validation set based on a ratio of 7:3. Univariate logistic regression analysis was utilized for the comparison of variables in training set. Candidate variables with P < 0.1 in training set was entered into the best subset selection, LASSO regression and Boruta feature selection. Finally, the selected variables significantly associated with the 3-year, 5-year, and 8-year OS were used to build a nomogram, followed by validation using receiver operating characteristic (ROC) curve, area under the curve (AUC), and calibration curve. Results: A total of 3265 rectal cancer patients (training set: 2285; test set: 980) were included in the present study. A nomogram was developed to predict the 3-year, 5-year, and 8-year OS based on age, household income, total number of in situ/malignant tumors, CEA, T stage, N stage and perineural invasion. The nomogram showed good efficiency in predicting the 3-year, 5-year and 8-year OS with good AUC for the training set and test set, respectively. Conclusion: We established a nomogram for predicting the 3-year, 5-year, and 8-year OS of the rectal cancer patients, which showed good prediction efficiency for the OS after TNT. [ABSTRACT FROM AUTHOR]

Due to the dynamic nature and limited resources in wireless networks, attack occurrence is inevitable. These attacks can damage or weaken the transmitted packets and threaten the entire system's efficiency. As a result, in such a situation, great and sometimes irreparable damage will be done to the business. Thus, security and attack prevention in wireless networks become a necessity and are very important. Essence intrusion detection systems determine whether a user's performance and behavior under the control or activity of a network traffic load is malicious. Since the characteristics of user behavior and network traffic are diverse and numerous, Selecting some features is necessary to improve the classification accuracy. Therefore, in this idea, a new model for estimating the penetration of wireless network-based networks is proposed based on a combination of feature subset selection based on firewall algorithm and fast neural learning networks. In this paper, the proposed idea will use the training set from the data set collected to test intrusion detection systems called KDD Cup to determine network intrusion detection methods and evaluate the proposed model. The proposed idea, based on the results obtained from the simulation and its performance in various experiments, has shown that it has improved significantly in terms of multiple criteria such as accuracy, F-criterion rate, and efficiency compared to the neural network pattern. In other words, the proposed idea performs better than the neural network method in identifying healthy nodes and new malicious intrusions in the target network. The simulation outputs also indicate that the proposed idea has a better classification rate and F-criteria than the FLN methods based on HSO, ATLBO, GA, and PSO. Vector backup machine, multilayer perceptron network, DBN, and S-NDAE have less time. [ABSTRACT FROM AUTHOR]

Selecting appropriate Global Climate Models (GCMs) presents a significant challenge for accurate climate projections. To address this, a novel framework based on information theory based minimum redundancy and maximum relevancy (MRMR) method identifies top-performing GCMs across the entire study region using multicriteria decision analysis methodology. A subset of the ten best-performing models out of twenty-two GCMs is chosen for multi-model ensemble analysis. Five MME methods are selected to assess the ensemble performance of the ten selected GCMs, categorized into simple, regression-based, geometric-based, and machine learning ensembles. This study evaluates the effectiveness of the MME method based on a comprehensive index called the extended distance between indices of simulation and observation. An Adaptive Multimodel Standardized Drought Index (AMSDI) has been developed based on the optimal MME method. For the application of the framework and the proposed index, historical precipitation data from 1950 to 2014 were utilized from 28 grid points in the Punjab province of Pakistan as the reference dataset. Additionally, simulations from 22 models of the Coupled Model Intercomparison Project phase 6, both past and future, were employed for the estimation procedure. In AMSDI indicator, we used improved multimodel ensemble of precipitation for future drought characterization under various future scenarios. Outcome associated with this research show that AMSDI effectively have ability to effectively identifiy extreme drought events for all three future scenarios. In conclusion, the AMSDI method is shown to be effective and flexible, improving accuracy in monitoring droughts. [ABSTRACT FROM AUTHOR]

Based on an analysis of the information processing mechanism in the primary visual cortex of biological vision, this study proposes an integration method of bar-combination of shifted filter responses (B-COSFIRE) filter with the differential evolution (DE) algorithm for enhancing the precision of events extraction. First, the B-COSFIRE filter incorporates trainable and unsupervised features, utilizing the two-dimensional expression of the difference-of-Gaussians (DoG) model to simulate the receptive field model. By capitalizing on the blur and shift properties of the DoG response, the proposed approach enhances the continuous effective signal while attenuating discontinuous noise signal, thereby demonstrating superior noise robustness compared to conventional methods. Second, the selectivity of proposed filter is not predefined during the implementation but automatically determined based on the given prototype pattern during the configuration process, resulting in a universal solution adaptable to various target patterns. Lastly, we employ the DE algorithm to optimize the feature selection process, enabling the extraction of a minimum feature subset that maximizes the performance of events characterization. The B-COSFIRE method is widely used in the field of image processing. When applying it to seismic exploration, the seismic data used by this algorithm is in 'sgy' format, providing richer information than traditional image data. The proposed model can effectively detect the event in seismic data with significant data volume and substantial noise interference. The B-COSFIRE filter method outperforms conventional edge detection techniques by accurately capturing seismic events of varying widths, aligning with the principles observed in biological vision mechanisms. The extracted events exhibit enhanced continuity and accuracy compared to existing approaches. [ABSTRACT FROM AUTHOR]

Cloud computing (CC) was extremely implemented by the application service providers (ASP) and enterprises for reducing either operational or capital expenditures. Services and applications before running on remote data centres can presently be transferred to public or private clouds. This alteration can manage the study community to analyze cloud platforms. But, a wide acceptance of such platforms can be blocked by main security concerns. Firewalls and typical rule-based security protection approaches could not be adequate for protecting user information in cloud environments. In recent times, advances in machine learning (ML) approaches are concerned the attention of research communities for building intrusion detection systems (IDS) that anomaly detection from the network traffic. This manuscript presents a Feature Subset Selection with Deer Hunting Optimizer based Deep Learning (FSS-DHODLAD) technique for Anomaly Detection in Secure Cloud Computing Environment. The purpose of the FSS-DHODLAD algorithm lies in the accurate identification and classification of anomalies that exist in the cloud platform. To reduce the high dimensionality of features, the FSS-DHODLAD technique designs a grasshopper optimization algorithm (GOA) to elect features. For anomaly detection, the FSS-DHODLAD technique utilizes Attention Convolutional Bidirectional Long Short Term Memory (AC-BLSTM) system. Eventually, the DHO system was utilized for the optimal hyperparameter selection of the AC-BLSTM model. To exhibit the higher performance of the FSS-DHODLAD system, a sequence of simulations was applied to the benchmark database and the outcomes are assessed concerning several evaluation measures. The simulation outcomes implied the greater performance of the FSS-DHODLAD algorithm over other recent approaches in terms of different metrics. [ABSTRACT FROM AUTHOR]

Tumor is one of the deadliest diseases; nowadays the cases of tumors are increasing rapidly. Researchers from worldwide are doing extensive research for the diagnosis and discernment of tumors by employing machine learning algorithms and performing experiments on the basis of observations which are stored in the form of datasets. The tumor-related dataset is high-dimensional and has many genes, most of which are not prognostic. Some of them are irrelevant and redundant. Here, we proposed a methodology named IG based on IWSSr-Random Forest(RF) which selects the most relevant prognostic genes by using Information Gain for gene ranking and evaluates the importance of genes by using RF after selecting the genes in an incremental manner in a wrapper. Furthermore, we use the RF for classification purposes. Experiments are performed on nine publicly available tumor-related datasets. Accuracy, Confusion matrix, Precision, F-measure, and Recall are used as performance evaluators. The proposed methodology selects 3 most relevant genes out of 2000 genes, 5 genes out of 7129 genes, 3 genes out of 7129 genes, 5 genes out of 24,481 genes, 7 genes out of 12,601 genes, 5 genes out of 15,154 genes, 2 genes out of 4026 genes, 5 genes out of 12,582 genes and 4 genes out of 2308 genes, and produces 88.71%, 71.67%, 98.61%, 79.38%, 93.60%, 99.60%, 92.42%, 95.83% and 92.77% accurate results in case of Colon, Central Nervous System, Leukemia, Breast Cancer, Lung Cancer, Ovarian Cancer, Lymphoma, MLL and SRBCT respectively. Experimental results present that IG based on IWSSr(RF) performs well compared to the state-of-the-art algorithms' results for instance RF, Naïve Bayes, KNN, and Decision Tree. IG based on IWSSr(RF) also has nominal time complexity compared to the time complexity of the above-mentioned classification algorithms. [ABSTRACT FROM AUTHOR]

Rising temperatures due to global warming necessitate immediate evaluation of heatwave patterns in Peninsular Malaysia (PM). For this purpose, this study utilized a locally developed heatwave index and a gridded daily maximum temperature (Tmax) dataset from ERA5 (1950–2022). During validation, the ERA5 dataset accurately represented the spatial pattern of Level 1 heatwaves, showing widespread occurrence. Historically, Level 1 heatwaves prevailed at 63.0%, followed by Level 2 at 27.7%, concentrated in northwestern states and the enclave between the Tahan and Titiwangsa mountain ranges. During very strong El Niño events in 1982/83, 1997/98, and 2015/16, Level 2 heatwave distributions were 10.4%, 26.8%, and 15.0%, respectively. For future projection, the model ensemble was created by selecting top-performing Global Climate Models (GCMs) using Kling-Gupta efficiency (KGE), ranked re-aggregation with compromise programming index (CPI), and GCM subset selection via Fisher-Jenks. The linear scaling bias-corrected GCMs (BC-GCMs), NorESM2-LM, ACCESS-CM2, MPI-ESM1-2-LR, ACCESS-ESM1-5, and FGOALS-g3, were found to exhibit better performance, and then ensemble. March to May show the highest increase in all scenarios, ranging from 3.3 °C to 4.4 °C for Level 1 heatwaves and 4.1 °C to 10.7 °C for Level 2 heatwaves. In the near future, SSP5-8.5 projects up to a 40.5% spatial increase for Level 1 heatwaves and a 2.3% increase for Level 2 heatwaves, affecting 97.1% and 57.2% of the area, respectively. In the far future, under SSP2-4.5 and SSP5-8.5, Tmax is projected to rise rapidly (1.5–4.5 °C) in the northern, western, and central regions, with increasing population exposure anticipated in the northern and western regions. [ABSTRACT FROM AUTHOR]

Variable selection involves selecting truly important predictors from p-dimensional multivariate functional predictors in functional predictive models. In this paper, a variable selection method is designed for scalar-on-function predictions entangled with nonlinear joint associations among scalar response and multiple functional predictors. First, a nonparametric functional nonlinear conditional correlation coefficient, namely, the FunNCC coefficient, is proposed to measure complex dependencies, including the nonmonotonic marginal dependence, along with the conditional associations of redundancy, complement, and interaction. Then, a model-free feature ordering and selection method is designed, where the FunNCC is utilized to rank relevance, enabling the selection of a subset of predictors with the strongest joint dependence. Since this method allows for quantitatively evaluating the contributions of predictors in explaining responses, it achieves moderate model interpretability. Finally, extensive simulation studies and two real-data cases involving air pollution regression and hand gesture recognition are conducted to evaluate the finite sample performance of the proposed method, and the results show that the proposed FunNCC and variable selection methods outperform state-of-the-art baselines. [ABSTRACT FROM AUTHOR]

Parkinson's disease (PD) arises from brain cell damage and necessitates early detection for effective treatment and symptom management. While various methods such as voice, speech, and written exams have been explored, utilizing automated tools is crucial to enhance accuracy. Recent advancements in artificial intelligence (AI) and deep learning (DL) provide an opportunity for precise early-stage PD identification. This study introduces a novel approach known as Quantum Mayfly Optimization-based feature subset selection with hybrid convolutional neural network (QMFOFS-HCNN) to improve PD detection and classification. QMFOFS-HCNN is designed to identify optimal feature subsets and overcome the dimensionality challenge. It combines a quantum mayfly optimization approach for feature selection with a convolutional neural network with attention-based long short-term memory for PD detection and classification. Additionally, hyperparameter selection is optimized using the Nadam optimizer. Experimental validation using benchmark datasets yielded compelling results. The QMFOFS-HCNN technique achieved accuracy rates: 96.35% for HandPD Spiral, 96.7% for HandPD Meander, 98.5% for Speech PD, and a perfect 100% for Voice PD datasets. These quantitative findings underscore the potential of AI and DL to enhance early PD detection accuracy significantly. These results offer promising prospects for improving healthcare outcomes in managing PD and related neurological disorders. [ABSTRACT FROM AUTHOR]

Machine learning models are revolutionizing our approaches to discovering and designing bioactive peptides. These models often need protein structure awareness, as they heavily rely on sequential data. The models excel at identifying sequences of a particular biological nature or activity, but they frequently fail to comprehend their intricate mechanism(s) of action. To solve two problems at once, we studied the mechanisms of action and structural landscape of antimicrobial peptides as (i) membrane-disrupting peptides, (ii) membrane-penetrating peptides, and (iii) protein-binding peptides. By analyzing critical features such as dipeptides and physicochemical descriptors, we developed models with high accuracy (86–88%) in predicting these categories. However, our initial models (1.0 and 2.0) exhibited a bias towards α-helical and coiled structures, influencing predictions. To address this structural bias, we implemented subset selection and data reduction strategies. The former gave three structure-specific models for peptides likely to fold into α-helices (models 1.1 and 2.1), coils (1.3 and 2.3), or mixed structures (1.4 and 2.4). The latter depleted over-represented structures, leading to structure-agnostic predictors 1.5 and 2.5. Additionally, our research highlights the sensitivity of important features to different structure classes across models. [ABSTRACT FROM AUTHOR]

Feature selection is one of the major components of the data processing flow that correctly selects real-time entities for categorization. It is utilized in numerous research fields, including the machine learning approach (Med Biol Eng Comput (Springer) 59(02):333–353 [1]). Feature selection and classification are beneficial to the processing of biomedical data in high-importance, high-dimensional datasets. Due to numerous obstacles in research, the current implementations are inadequate for predicting classification accuracy. To handle the classification challenge, we require dedicated neural network and classification model approaches (J Comput Mater Continua (CMC) 72(1):243–259 [2]). Initial features are selected using a method called "fuzzy c-means clustering with rough set theory" and are subsequently classified using "support vector machines." In addition, the current approach is extremely time-consuming. The proposed solution used efficient feature subset selection in high-dimensional data to fix these problems. To begin with, our recommended method used Enhanced Social Spider Optimization (ESSO) computation to select the best highlights (IEEE Trans Circuits Syst Video Technol (TCSVT) 28(2):454–467 [3]). When categorizing data, the Optimal Radial Basis Function Neural Network (ORBFNN) is used. Methods for calculating Artificial Bee Colony (ABC) are used to streamline RBFNN's sufficiency in characterizing smaller scale show information. [ABSTRACT FROM AUTHOR]

Numerous studies emphasize accuracy in machine learning regression models, yet scalability and execution efficiency are often overlooked, critical for large datasets or extensive computations. This paper introduces a scalable, distributed Spark MLlib regression model through the best subset selection approach to predict Covid-19 statistics in India, demonstrating high accuracy, scalability, and execution efficiency. Notably, limited research focuses on tree-based regression, particularly gradient boost regression, in the context of the Covid-19 dataset. The proposed work optimizes regression models for accuracy and execution time on Spark clusters of varying sizes using the best subset selection approach. Evaluation encompasses Root Mean Square Error (RMSE), Mean Absolute Error (MAE), R2 Error for accuracy, and execution time analysis. Results indicate superior prediction accuracy in tree-based regression, with Gradient Boosted Tree Regression (GBTR) leading, and Random Forest Regression (RFR) surpassing Decision Tree Regression (DTR). Accuracy remains consistent across Python library, Spark MLlib on a single machine, and clusters of varying sizes, with Spark MLlib displaying lower execution times than Python's machine learning library on a single machine. Furthermore, execution times decrease substantially within Spark clusters, particularly for the iterative GBTR. This research uncovers scalability and execution efficiency aspects, highlighting tree-based regression's accuracy and advocating for Spark MLlib's efficacy in enhancing execution efficiency, especially across multi-node clusters. [ABSTRACT FROM AUTHOR]

The visual system can rapidly calculate the ensemble statistics of a set of objects; for example, people can easily estimate an average size of apples on a tree. To accomplish this, it is not always useful to summarize all the visual information. If there are various types of objects, the visual system should select a relevant subset: only apples, not leaves and branches. Here, we ask what kind of visual information makes a "good" ensemble that can be selectively attended to provide an accurate summary estimate. We tested three candidate representations: basic features, preattentive object files, and full-fledged bound objects. In four experiments, we presented a target and several distractors' sets of differently colored objects. We found that conditions where a target ensemble had at least one unique color (basic feature) provided ensemble averaging performance comparable to the baseline displays without distractors. When the target subset was defined as a conjunction of two colors or color-shape partly shared with distractors (so that they could be differentiated only as preattentive object files), subset averaging was also possible but less accurate than in the baseline and feature conditions. Finally, performance was very poor when the target subset was defined by an exact feature relationship, such as in the spatial conjunction of two colors (spatially bound object). Overall, these results suggest that distinguishable features and, to a lesser degree, preattentive object files can serve as the representational basis of ensemble selection, while bound objects cannot. [ABSTRACT FROM AUTHOR]

High dimensionality can lead to overfitting and affect the modeling power of classification algorithms, resulting an increase in false positive rate (FPR) and false negative rate (FNR). Therefore, feature selection is a critical issue to deal with by means of efficient techniques when developing Intrusion Detection Systems. This study seeks to assess and compare the impacts of five univariate filters (Relieff, Pearson Correlation, Mutual Information, ANOVA, and Chi2) with different selection thresholds, and three multivariate filters (Correlation-based feature subset selection, Double Input Symmetric Relevance and Consistency-based subset selection) on the performances of four classifiers (Multilayer Perceptron, Support Vector Machines, XGBoost and Random Forest) over CIC-IDS2017, CSE-CIC-IDS2018 and CIC-ToN-IoT intrusion detection datasets. We evaluate 228 variants of classifiers to determine the features that positively impact the classification efficiency of all the cyber-attack scenarios used. The obtained results show that using XGBoost and Random Forest trained with multivariate methods, such as CON and DISR, can effectively reduce the number of features without affecting the classification performance and detection rate, compared to other filtering methods. [ABSTRACT FROM AUTHOR]

The interpretation of complex biological datasets requires the identification of representative variables that describe the data without critical information loss. This is particularly important in the analysis of large phenotypic datasets (phenomics). Here we introduce Multi-Attribute Subset Selection (MASS), an algorithm which separates a matrix of phenotypes (e.g., yield across microbial species and environmental conditions) into predictor and response sets of conditions. Using mixed integer linear programming, MASS expresses the response conditions as a linear combination of the predictor conditions, while simultaneously searching for the optimally descriptive set of predictors. We apply the algorithm to three microbial datasets and identify environmental conditions that predict phenotypes under other conditions, providing biologically interpretable axes for strain discrimination. MASS could be used to reduce the number of experiments needed to identify species or to map their metabolic capabilities. The generality of the algorithm allows addressing subset selection problems in areas beyond biology. Multi-Attribute Subset Selection is an algorithm for identifying the most descriptive growth conditions in microbial phenotyping experiments. This streamlines high-throughput phenotyping efforts and helps to reveal biological insights. [ABSTRACT FROM AUTHOR]

Sensing and subsequent analysis of the environmental data of a given geographical area is an essential requisite for the planned development of that region. Nowadays, IoT Sensor-Cloud ecosystem has been adopted to collect data from IoT sensors and transmit it to the chosen Cloud Server for further processing and dissemination. In a large Wireless Sensor Network formed by the IoT sensors, there will be a significant amount of redundancy in the dataset when the nodes are placed closely, and the sensed data varies slowly and gradually with regard to time and space. Then, avoiding redundant data transmission can lead to lower energy consumption and communication overhead. Adaptive subset selection of sensor nodes for data size reduction in a Wireless Sensor Network is an approach to efficiently managing the amount of data transmitted within the network. Then, in the current time schedule, it is possible to optimally select a subset of the sensor nodes for data collection without very much affecting the overall data fidelity. An optimal sensor node subset selection scheme that reduces the communication load with minimum information loss is proposed to achieve this task. The unselected nodes are put in sleep mode, which consequently results in lower sensor energy expenditure. The subset selection algorithm is implemented based on the derivative-free pattern search optimizer that minimizes the reconstruction error during the associated extrapolation. This approach differs entirely from the Compressive Data Gathering approach. The simulation results reveal that the performance of the proposed scheme is superior to other similar competitive methods in terms of the mean square error, which is found to be 1.95, with the percentage participation nodes equal to 50% and when the sensor data is uniformly distributed over 20 and 30 units. [ABSTRACT FROM AUTHOR]

Adverse Outcome Pathway (AOP) is a useful tool to glean mode of action (MOE) of a chemical. However, in order to use it for the purpose of risk assessment, an AOP needs to be quantified using in vitro or in vivo data. Majority of quantitative AOPs developed so far, were for single exposure to progressively higher doses. Limited attempts were made to include time in the modeling. Here as a proof-of concept, we developed a hypothetical AOP, and quantified it using a virtual dataset for six repeated exposures using a Bayesian Network Analysis (BN) framework. The virtual data was generated using realistic assumptions. Effects of each exposure were analyzed separately using a static BN model and analyzed in combination using a dynamic BN (DBN) model. Our work shows that the DBN model can be used to calculate the probability of adverse outcome when other upstream KEs were observed earlier. These probabilities can help in identification of early indicators of AO. In addition, we also developed a data driven AOP pruning technique using a lasso-based subset selection, and show that the causal structure of AOP is itself dynamic and changes over time. This proof-of-concept study revealed the possibility for expanding the applicability of the AOP framework to incorporate biological dynamism in toxicity appearance by repeated insults. [ABSTRACT FROM AUTHOR]

Physiologically-based pharmacokinetic (PBPK) modeling is important for studying drug delivery in the central nervous system, including determining antibody exposure, predicting chemical concentrations at target locations, and ensuring accurate dosages. The complexity of PBPK models, involving many variables and parameters, requires a consideration of parameter identifiability; i.e., which parameters can be uniquely determined from data for a specified set of concentrations. We introduce the use of a local sensitivity-based parameter subset selection algorithm in the context of a minimal PBPK (mPBPK) model of the brain for antibody therapeutics. This algorithm is augmented by verification techniques, based on response distributions and energy statistics, to provide a systematic and robust technique to determine identifiable parameter subsets in a PBPK model across a specified time domain of interest. The accuracy of our approach is evaluated for three key concentrations in the mPBPK model for plasma, brain interstitial fluid and brain cerebrospinal fluid. The determination of accurate identifiable parameter subsets is important for model reduction and uncertainty quantification for PBPK models. [ABSTRACT FROM AUTHOR]

Since either only one metric is employed or two metrics are adopted and each of them is compared separately to measure redundant features, several existing feature subset selection algorithms cannot obtain the desired performance. To address the problem, a feature subset selection algorithm named symmetric uncertainty and interaction factor (SUIF) is presented. SUIF first exploits symmetric uncertainty to evaluate relevant features and removes irrelevant features. Then, it uses a graph theoretic representation to process these relevant features and removes some edges whose weights are some smaller values. Following that, it adopts Louvain community detection algorithm to cluster features into several clusters. Finally, it utilizes two metrics, symmetric uncertainty and interaction factor, and the method of equal interval division and ranking to evaluate features in each cluster, and removes redundant features. For validating the performance of SUIF, FCBF, INTERACT, FAST, SAOLA, NFCBF and EID-TII are exploited for comparison. Experimental results indicate that SUIF can obtain better feature selection performance. [ABSTRACT FROM AUTHOR]

Since the last decade, high-dimensional data has been increasing in various document mining fields, such as text summarization, text clustering, and text classification. The curse of dimensionality has an impact on the classification model's performance. The feature selection strategy is extremely effective in dealing with the curse of dimensionality issue. In this work, we present the Tom and Jerry Optimization technique(TJO) for feature subset selection. The proposed work uses the classifier error rate and the feature chosen rate to measure the candidate's fitness. The performance of the proposed scheme is examined using two popular benchmark text corpus and compared with five metaheuristic approaches. The best success rate obtained by the proposed scheme is 95.77%, whereas the best precision is 0.9509, recall is 0.9577 and F1-score is 0.9541. According to the comparison results, the proposed feature subset selection scheme outperforms the standard strategy. [ABSTRACT FROM AUTHOR]

Multimodal multi-objective optimization problems (MMOPs) are widely present in real life. Due to the need of balancing between convergence and diversity in multi-objective optimization, as well as the need of balancing between diversities in objective and decision spaces, exploring the Pareto optimal front and Pareto optimal solution set becomes rather difficult in solving MMOPs. Recently, some multimodal multi-objective optimization algorithms have emerged. However, most of them are convergence-first, which may result in poor diversity of the solution set in decision space. To remedy this defect, in this paper, a determinantal point process (DPP)-assisted evolutionary algorithm is proposed to effectively solve MMOPs. In the proposed method, i) the DPPs are used to select subsets to consider convergence and diversity in both objective and decision spaces; ii) a kernel matrix is designed to retain solutions with poor convergence but good diversity in the decision space to explore the equivalent Pareto optimal solution sets; and iii) we propose a framework that combines the population and archive to better solve MMOPs. The results show that the proposed algorithm achieves the best performance in 18 of the 28 benchmark problems compared to six state-of-the-art algorithms. [ABSTRACT FROM AUTHOR]

In the era of big data, the demand for multivariate time series prediction has surged, drawing increased attention to feature selection and neural networks in machine learning. However, certain feature selection methods neglect the alignment between actual data sample differences and clustering results, while neural networks lack automatic parameter adjustment in response to changing target features. This paper presents the MBSSA-Bi-AESN model, a Bi-directional Adaptive Echo State Network that utilizes the modified salp swarm algorithm (MBSSA) and feature selection to address the limitations of manually set parameters. Initial feature subset selection involves assigning weights based on the consistency of clustering results with differences. Subsequently, the four critical parameters in the Bi-AESN model are optimized using MBSSA. The optimized Bi-AESN model and selected feature subset are then integrated for simultaneous model learning and optimal feature subset selection. Experimental analysis on eight datasets demonstrates the superior prediction accuracy of the MBSSA-Bi-AESN model compared to benchmark models, underscoring its feasibility, validity, and universality. [ABSTRACT FROM AUTHOR]

Market analyzers use different parameters as features in the market data to analyze the market trends. The feature's values act as a signal to market fluctuations. Many studies have examined these features to predict market movement more effectively. However, the method to minimize the uncertainties associated with the features is not available in the literature. This exploratory study introduces the uncertainty optimization based feature selection method for stock marketing. We introduce a notion of certainty region of the feature as the set of feature values, which signify particular happening with certainty. We use rough set theory to find the feature's certainty region and uncertainty region and measure each feature's significance. The feature whose certainty region is the maximum is the most significant in the feature space. Hence we group the features by minimizing the uncertainty region of the most informative features to get feature subsets for feature selection. We propose an algorithm based on uncertainty optimization to find subsets of the feature set for effectiveness and performance enhancement in the feature selection. We obtain the decision rules with comprehensive coverage and excellent support using the selected features. The accuracy of classification using the chosen parameters is up to 85.91%, which is higher than 79.54% of the complete feature set. The study provides an uncertainty optimization model for more efficient market movement prediction. [ABSTRACT FROM AUTHOR]

We conduct factor importance analysis on the insurance transitioning behavior of the public health insurance enrollees in the U.S., based on the Medical Expenditure Panel Survey (2-year longitudinal files, 2011–2018). The factors in the dataset consist of both numerical and categorical type data, which makes it very challenging to compare their importances. In order to measure the effects of these factors of mixed data types, two relatively "model-free" ranking scores are introduced: vote count and average partial dependence ranking score. Different from traditional ranking scores, the novel ranking scores apply to any regression and classification methods and to both numerical and categorical factors. A voting ensemble is designed to obtain the above two ranking scores, with four competitive base learners: forward and backward stepwise subset selections, LASSO and random forest. The top five driving factors selected by our voting ensemble are number of physician office visits, family size, chronic condition, age and family income. A predictive model based on the top-ranked factors is provided. This model is competitive to other popular prediction methods, according to our model validation result. [ABSTRACT FROM AUTHOR]

This work presents a novel approach for optimal feature subset (OFS) selection in weather prediction (WP), addressing the challenge of handling a large number of features. The proposed method is a filter-based technique utilizing an intuitionistic fuzzy inference system (IFIS) designed to assess relationships between meteorological features while incorporating geographical factors. The core focus is on the utilisation of the 'hesitation degree' (HD) as a measure of feature importance, a concept applied for the first time in this domain. The method is compared against traditional and state-of-the-art algorithms, including custom fuzzy inference systems (FIS) and several variations of IFIS, showcasing its superiority in terms of accuracy (ACC), precision (PRE), recall (REC), and f1-score (F1S) across various classifiers. The computational analysis affirms the simplicity and efficiency of the proposed method. The main contributions encompass the development of a computationally efficient filter-based feature selection (FS) method, the integration of geographical features, and the emphasis on the HD for a nuanced FS, demonstrating robust performance in scenarios involving nonlinear relationships between features and the target feature. [ABSTRACT FROM AUTHOR]

We show a connection between lower and upper bounds on the column matrix approximation accuracy and the bounds on the norms of the pseudoinverses of the submatrices of orthogonal matrices. This connection is exploited to derive lower bounds for column approximations accuracy in spectral and Frobenius norms. [ABSTRACT FROM AUTHOR]

We consider the problem of subset selection for ℓ p subspace approximation and (k, p)-clustering. Our aim is to efficiently find a small subset of data points such that solving the problem optimally for this subset gives a good approximation to solving the problem optimally for the original input. For ℓ p subspace approximation, previously known subset selection algorithms based on volume sampling and adaptive sampling proposed in Deshpande and Varadarajan (STOC'07, 2007), for the general case of p ∈ [ 1 , ∞) , require multiple passes over the data. In this paper, we give a one-pass subset selection with an additive approximation guarantee for ℓ p subspace approximation, for any p ∈ [ 1 , ∞) . Earlier subset selection algorithms that give a one-pass multiplicative (1 + ϵ) approximation work under the special cases. Cohen et al. (SODA'17, 2017) gives a one-pass subset section that offers multiplicative (1 + ϵ) approximation guarantee for the special case of ℓ 2 subspace approximation. Mahabadi et al. (STOC'20, 2020) gives a one-pass noisy subset selection with (1 + ϵ) approximation guarantee for ℓ p subspace approximation when p ∈ { 1 , 2 } . Our subset selection algorithm gives a weaker, additive approximation guarantee, but it works for any p ∈ [ 1 , ∞) . We also focus on (k, p)-clustering, where the task is to group the data points into k clusters such that the sum of distances from points to cluster centers (raised to the power p) is minimized for p ∈ [ 1 , ∞) . The subset selection algorithms are based on D p sampling due to Wei (NIPS'16, 2016) which is an extension of D 2 sampling proposed in Arthur and Vassilvitskii (SODA'07, 2007). Due to the inherently adaptive nature of the D p sampling, these algorithms require taking multiple passes over the input. In this work, we suggest one pass subset selection for (k, p)-clustering that gives constant factor approximation with respect to the optimal solution with an additive approximation guarantee. Bachem et al. (NIPS'16, 2016) also gives one pass subset selection for k-means for p = 2 ; however, our result gives a solution for a more generic problem when p ∈ [ 1 , ∞) . At the core, our contribution lies in showing a one-pass MCMC-based subset selection algorithm such that its cost incurred due to the sampled points closely approximates the corresponding optimal cost, with high probability. [ABSTRACT FROM AUTHOR]

Real-world problems are commonly characterized by a high feature dimensionality, which hinders the modelling and descriptive analysis of the data. However, some of these data may be irrelevant or redundant for the learning process. Different approaches can be used to reduce this information, improving not only the speed of building models but also their performance and interpretability. In this review, we focus on feature subset selection (FSS) techniques, which select a subset of the original feature set without making any transformation on the attributes. Traditional batch FSS algorithms may not be adequate to efficiently handle large volumes of data, either because memory problems arise or data are received in a sequential manner. Thus, this article aims to survey the state of the art of incremental FSS algorithms, which can perform more efficiently under these circumstances. Different strategies are described, such as incrementally updating feature weights, applying information theory or using rough set-based FSS, as well as multiple supervised and unsupervised learning tasks where the application of FSS is interesting. [ABSTRACT FROM AUTHOR]

Various fuzzy covering based rough set models can characterize and approximate a given target concept with different lower and upper approximation operators. However, due to the lack of effective construction method and semantic interpretation on real data, there are few investigation on a comprehensive uncertain measure for fuzzy covering, as well as its applications in feature subset selection. For this reason, a new uncertain measure called multi-level granularity entropy is proposed by means of different fuzzy neighbourhoods. First, five types of granularity structure and granularity entropy for fuzzy covering are studied; Second, a multi-level granularity entropy is introduced, using which one can evaluate the discrimination information implied in a family of fuzzy coverings from different hierarchical levels. Some variants of the multi-level granularity entropy, such as conditional granularity entropy, joint granularity entropy, and mutual granularity entropy, are then proposed to reflect the change of uncertain information. Finally, the dimensionality reduction is implemented in the light of maintaining the discrimination ability, and a forward feature selection algorithm is developed by using the multi-level entropy. Extensive numerical experiments demonstrate that the presented model exhibits better performance in feature learning, and outperforms some representative feature selection algorithms in terms of classification accuracy and the number of selected features. [ABSTRACT FROM AUTHOR]

Ensemble pruning can improve the performance and reduce the storage requirements of an integration system. Most ensemble pruning approaches remove low-quality or redundant classifiers by evaluating the classifiers' competence and relationships via their predictions. However, finding the best way to represent classifiers and create ensemble diversity is still a worthy research problem in the ensemble pruning field. To confront this issue, we discuss whether properties other than predictions can represent classifiers and propose a new classifier selection method, classifier-representation- and clustering-ensemble-based ensemble pruning (CRCEEP). In the proposed method, two new classifier-representation-learning methods, local-space- and relative-transformation-based representation, are proposed to obtain more information about classifiers. CRCEEP incorporates the clustering ensemble method to group classifiers and prune redundant learners. Finally, accurate and diverse classifiers are integrated to improve classification performance. Extensive experiments were carried out on UCI datasets, and the experimental results verify CRCEEP's effectiveness and the necessity of classifier representation. [ABSTRACT FROM AUTHOR]

Micro-ribonucleic acids (miRNAs) are tiny noncoding ribonucleic acid (RNA) molecules that involve various biological processes for cancer advancements. Classification of cancer is a crucial problem in microarray technology because of the huge number of miRNAs and a limited number of samples. Therefore, to improve the classification of cancer between benign and malignant samples, this paper presents a hybrid filter-wrapper method that uses the multi-filter ensemble (MFE) approach as a filter method used in selecting the top-ranked miRNAs. For the wrapper approach, three different algorithms based on simultaneous utilization of the grey wolf optimizer (GWO), evolutionary population dynamics (EPD), and two selection mechanisms, are presented to enhance the GWO algorithm performance. The first approach involves the application of the EPD mechanisms to the GWO to eliminate the worst solutions of GWO and repositions them toward alpha, beta, or delta wolves to improve exploitation. So also, the GWO needed to randomly reinitialize its poor solutions near to the search space, using the EPD for the enhancement of the exploration process. In the second and third approaches, the Roulette wheel selection (RWS) and tournament selection (TS) are applied to offer a chance for low fitness individuals to be selected during the search process that maintains the diversity of the selected solutions. The proposed GWO_EPD algorithms are applied for miRNAs selection, and obtained results demonstrate that the EPD can improve the search capability of GWO in terms of avoiding local optima, exploration, and convergence rate. Moreover, the results of GWO_EPD are compared with twelve miRNAs subset selection approaches using classification accuracy and the number of selected miRNAs. [ABSTRACT FROM AUTHOR]

Simplicity, efficiency and precision are basic principles for modeling and analyzing of coal-fired boilers data. However, the load fluctuations, system delay, multi-variable coupling pose great challenges to the high-precision modeling for NOx emission. A hybrid feature selection method for boiler data is proposed to predict accurately NOx emission. Firstly, the mechanism analysis is used to narrow the feature scope, and the feature set is selected preliminarily. Secondly, maximum information coefficient (MIC) method is introduced to calculate the correlation between features and NOx information to eliminate boiler system delay. Thirdly, a combined feature evaluation method is developed, which integrates Filter and Embedded method to obtain feature ranking, then the ranking information is regarded as priori knowledge to improve the genetic algorithm. Finally, a fitness function to maximize prediction accuracy and minimize feature dimension is constructed based on hybrid method to realize feature subset selection and NOx emission prediction. Original data are collected from one 1000 MW coal-fired unit in the Guangdong province of China. Experimental results show that the number of features is reduced by nearly 70%, the MAPE of LightGBM regression model is no more than 2%. Higher prediction accuracy can be obtained using the features extracted through the proposed hybrid method. [ABSTRACT FROM AUTHOR]

Glacier response patterns at the catchment scale are highly heterogeneous and defined by a complex interplay of various dynamics and surface factors. Previous studies have explained heterogeneous responses in qualitative ways but quantitative assessment is lacking yet where an intrazone homogeneous climate assumption can be valid. Hence, in the current study, the reason for heterogeneous mass balance has been explained in quantitative methods using a multiple linear regression model in the Sikkim Himalayan region. At first, the topographical parameters are selected from previously published studies, then the most significant topographical and geomorphological parameters are selected with backward stepwise subset selection methods. Finally, the contributions of selected parameters are calculated by least square methods. The results show that, the magnitude of mass balance lies between −0.003±0.24 to −1.029±0.24 m.w.e.a−1 between 2000 and 2020 in the Sikkim Himalaya region. Also, the study shows that, out of the terminus type of the glacier, glacier area, debris cover, ice-mixed debris, slope, aspect, mean elevation, and snout elevation of the glaciers, only the terminus type and mean elevation of the glacier are significantly altering the glacier mass balance in the Sikkim Himalayan region. Mathematically, the mass loss is approximately 0.40 m.w.e.a−1 higher in the lake-terminating glaciers compared to the land-terminating glaciers in the same elevation zone. On the other hand, a thousand meters mean elevation drop is associated with 0.179 m.w.e.a−1 of mass loss despite the terminus type of the glaciers. In the current study, the model using the terminus type of the glaciers and the mean elevation of the glaciers explains 76% of fluctuation of mass balance in the Sikkim Himalayan region. [ABSTRACT FROM AUTHOR]

Extant sequential wrapper-based feature subset selection (FSS) algorithms are not scalable and yield poor performance when applied to big datasets. Hence, to circumvent these challenges, we propose parallel and distributed hybrid evolutionary algorithms (EAs) based wrappers under Apache Spark. We propose two hybrid EAs based on the Binary Differential Evolution (BDE), and Binary Threshold Accepting (BTA), namely, (i) Parallel Binary Differential Evolution and Threshold Accepting (PB-DETA), where BDE and BTA work in tandem in every iteration, and (ii) its ablation variant, Parallel Binary Threshold Accepting and Differential Evolution (PB-TADE). Here, BTA is invoked to enhance the search capability and avoid premature convergence of BDE. For comparison purposes, we also parallelized two state-of-the-art algorithms: adaptive DE (ADE) and permutation based DE (DE-FSPM), and named them PB-ADE and P-DE-FSPM respectively. Throughout, logistic regression (LR) is employed to compute the fitness function, namely, area under the receiver operator characteristic curve (AUC). The effectiveness of the proposed algorithms is tested over the five big datasets of varying dimensions. It is noteworthy that the PB-TADE turned out to be statistically significant than the rest. All the algorithms have shown the repeatability property. The proposed parallel model attained a speedup of 2.2–2.9. We also reported feature subset with high AUC and least cardinality. [ABSTRACT FROM AUTHOR]

Detecting automated users' agent activities at any web application through users' web access logs is a challenging issue. Many machines learning based automated solutions exist to address this issue. However, the existing supervised learning methods are heavily dependent on fully labeled data. But the scarcity of labeled log access data and the cost of labeling still make the issue challenging. Some unsupervised learning-based solutions are also proposed, but their performance accuracy is questionable. The semi-supervised based self-training method works with a small set of partially labeled data but lacks a suitable selection metric for a set of predictions with a high degree of confidence and a reliable base learner. In this paper, we propose a new semi-supervised learning based self-training method using probability-based selection criteria with Mahalanobis distance, named DIstance-based SElf-Training (DISET) for detecting automated users' agent activities. The DISET used probability-based selection criteria with Mahalanobis distance to achieve high-confidence subset selection. The DISET framework works in four steps. First, it performs the data cleaning, session identification, feature extraction, and session labeling during the data preprocessing step. The second step segments the data into labeled and unlabeled datasets. The third step of model self-training performs the subset selection using six different supervised base learners independently. Lastly, the fourth step tests the performance of the used model. The performance of DISET is evaluated on NASA95 and E-commerce weblog datasets using three-fold cross-validation training and testing. The used datasets are also divided into different ratios of labeled and unlabeled instances for experiments. The performance is recorded on the accuracy, precision, recall, and the f-1 score, and the Matthews Correlation Coefficient (MCC) measures and compares the model's performance with six different base classifiers. We also plotted the ROC and PR curves to confirm and compare the performance of different base learners with the DISET method. Out of the six-base learners, XGBoost outperformed both datasets in the 30:70 data segmentation ratio. The results show that DISET achieves a minimum percentage improvement of 1.91% in accuracy, 2.70% in precision, 3.65% in sensitivity, and 1.00% in F-1 score with large unlabeled datasets. [ABSTRACT FROM AUTHOR]

Imbalance classification has received great attention due to its various real-world applications. Data-level approaches are the most convenient to address data imbalance, whereas oversampling is the most deeply explored. However, most previous studies used distance-based factors to select minority class instances for oversampling. Thus, the synthetic instances often did not follow the distribution of the original minority class instances. In this work, we propose a novel oversampling method based on instance hardness and multivariate Gaussian distribution. First, a fused feature set including k-disagree value and classification error is used for selecting and weighting minority class instances for oversampling. Here, the k-disagree value is also used to filter majority class instances. Then, multivariate Gaussian distribution is fitted to the subset of selected minority class instances, where the selection of subset is based on closest- and cluster-based methods. Next, new instances are generated based on the subset distribution. Finally, Euclidean distance-based instance selection is investigated for improved imbalance classification performance. Experimental results on the KEEL dataset repository show that our proposed method can outperform the other compared oversamplers in terms of both AUC and G-mean. [ABSTRACT FROM AUTHOR]

In order to improve the accuracy of crosslinked polyethylene (XLPE) cable status diagnosis and eliminate redundant features, this paper proposes a feature selection method for XLPE cable status diagnosis based on particle swarm optimization algorithm to optimize support vector machine. Firstly, this paper constructs 31 binary coded feature combinations based on five commonly used feature parameters for XLPE cable detection. Secondly, the optimal feature subset is selected based on the average accuracy of 20 cross-validation of PSO-SVM under 31 different feature combinations. Finally, train the diagnostic model with the optimal feature subset and its corresponding parameters, and verify the diagnostic performance of the selected features with test set samples and XLPE cable field data. The experimental results show that the PSO-SVM feature selection method can effectively find the optimal feature subset. The number of features selected in this paper is 2, and the classification accuracy of the test set is 98%. Compared with the preferred features of traditional feature selection methods, the features selected in this paper have good generalization ability. [ABSTRACT FROM AUTHOR]

Wetlands are one of the world's most significant and vulnerable ecosystems. The wetlands of the Yellow River Delta are subject to multiple influences of ocean tidal action and the massive sediment deposits of the Yellow River, resulting in a more complex and unstable composition of land cover types. To better distinguish the wetlands in the region, we conducted the classification using an object-oriented combined with feature preference machine learning approach. To alleviate the pretzel phenomenon in pixel-based classification, a superpixel segmentation method using the watershed algorithm with H-minima labeling was used to segment the images at the optimal scale. The best feature subset for classification was filtered using the recursive feature elimination cross-validation approach, which extracts multiple spectral indices from the images. A random forest classifier combining superpixel segmentation and feature selection methods was proposed for the wetland classification. The model improves the classification accuracy of wetlands compared to three classical pixel-based machine learning classification methods. And the overall accuracy was 91.74% and the kappa coefficient was 0.9078, both of which were improved by about 4.53% and 0.0506, respectively, compared with the best-performing random forest classifier in pixel-oriented. The results showed that this method can effectively improve the classification accuracy of the Yellow River Delta wetlands compared with the previous studies. [ABSTRACT FROM AUTHOR]

A bottleneck in Laser Powder Bed Fusion (L-PBF) metal additive manufacturing (AM) is the quality inconsistency of its products. To address this issue without costly experimentation, computational multi-physics modeling has been used, but the effectiveness is limited by parameter uncertainties and their interactions. We propose a full factorial design and variable selection approach for the analytics of main and interaction effects arising from material parameter uncertainties in multi-physics models. Data is collected from high-fidelity thermal-fluid simulations based on a 2-level full factorial design for 5 selected material parameters. Crucial physical phenomena of the L-PBF process are analyzed to extract physics-based domain knowledge, which are used to establish a validation checkpoint for our study. Initial data visualization with half-normal probability plots, interaction plots and standard deviation plots, is used to assess if the checkpoint is being met. We then apply the combination of best subset selection and the LASSO method on multiple linear regression models for comprehensive variable selection. Analytics yield statistically and phyiscally validated findings with practical implications, emphasizing the importance of parameter interactions under uncertainty, and their relation to the underlying physics of L-PBF. [ABSTRACT FROM AUTHOR]

This work proposes an efficient and promising method for linear and nonlinear dynamic systems modeling. Unlike previous methods, we combine Fuzzy C-Means (FCM) and modified Particle Swarm Optimization (PSO) algorithms. The FCM allows clustering large nonlinear data sets (good rule-base parameters initialization). The PSO algorithm is used to optimize the parameters of the fuzzy system rules (initialized by FCM). This combination allows reducing the rule-base of TSK fuzzy model considering the three goals: have a good initialization of the parameters to be optimized, handling complex systems with ensuring a high accuracy and a low complex algorithm structure. Indeed, from a given fuzzy rule-base, FCM–PSO selects a subset of important fuzzy rules based on the threshold criterion of each rule. The interaction between the particles improves iteratively the quality of each fuzzy rule. Simulation results using two well-known benchmark functions show the efficiency of the proposed approach when compared to previous works. [ABSTRACT FROM AUTHOR]

Research on low-sulfur and ultra-low-sulfur steel is an important field, since sulfur is known to increase hydrogen-induced cracking, lower hot ductility, and increase solidification cracking susceptibility. Several approaches can be used to assess the sulfur removal performance in steel-making plants, and to improve the desulfurization reaction. Statistical data-driven approaches improve the capacity to build mathematical models that merge all thermodynamics and kinetics features, and measure each one's importance in the specific studied process. The present study aims to model the desulfurization process from the steel tapping until the ladle arrival at the ladle furnace station in a billet steel-making plant. It was performed using Pearson correlation filters and an implementation of a best subset selection algorithm with multiple linear regression. It was observed that, for the analyzed heats, the silicon carbide added during the tapping, the ladle transfer time, the EAF slag viscosity, and the activity of CaO·Cr2O3 are the most important features that influence the desulfurization efficiency between the steel tapping and the arrival at the ladle furnace station. It was observed that the metallurgical theoretical background is an essential requirement to build appropriate process mathematical models and avoid wrong decisions. [ABSTRACT FROM AUTHOR]

The rate and extent of biodegradation of petroleum hydrocarbons in the different aquatic environments is an important element to address. The major avenue for removing petroleum hydrocarbons from the environment is thought to be biodegradation. The present study involves the development of predictive quantitative structure–property relationship (QSPR) models for the primary biodegradation half-life of petroleum hydrocarbons that may be used to forecast the biodegradation half-life of untested petroleum hydrocarbons within the established models' applicability domain. These models use easily computable two-dimensional (2D) descriptors to investigate important structural characteristics needed for the biodegradation of petroleum hydrocarbons in freshwater (dataset 1), temperate seawater (dataset 2), and arctic seawater (dataset 3). All the developed models follow OECD guidelines. We have used double cross-validation, best subset selection, and partial least squares tools for model development. In addition, the small dataset modeler tool has been successfully used for the dataset with very few compounds (dataset 3 with 17 compounds), where dataset division was not possible. The resultant models are robust, predictive, and mechanistically interpretable based on both internal and external validation metrics (R2 range of 0.605–0.959. Q2(Loo) range of 0.509–0.904, and Q2F1 range of 0.526–0.959). The intelligent consensus predictor tool has been used for the improvement of the prediction quality for test set compounds which provided superior outcomes to those from individual partial least squares models based on several metrics (Q2F1 = 0.808 and Q2F2 = 0.805 for dataset 1 in freshwater). Molecular size and hydrophilic factor for freshwater, frequency of two carbon atoms at topological distance 4 for temperate seawater, and electronegative atom count relative to size for arctic seawater were found to be the most significant descriptors responsible for the regulation of biodegradation half-life of petroleum hydrocarbons. [ABSTRACT FROM AUTHOR]

Optimal selection of features leads to the increase in perceptibility of the predictor procedure and thereby in turn increases the accuracy. In the domain of WiFi-based indoor positioning, access points are a ubiquitous source of features with which the positioning process is carried out. The selection of a proper subset of access points is crucial for sustainable localization performance. Most of the existing works in the literature focus on important AP selection whereas, the stability of the feature sets w.r.t varying context remains mostly ignored. Thus, to impart sustainable localization performance, in this work, the problem of identification of a stable set(s) of APs is investigated. Thus, the contribution of this work is two-fold. A genetic algorithm based feature selection technique for indoor localization is proposed first. The motive behind the approach is to capture different subsets of APs that best contribute to the localization process. Accordingly, our second contribution is the design of a Feature Association Based Ensemble Learning (FABEL) model that utilizes the selected feature sets in order to retain generality for sustainable localization performance. Experimentation has been carried out on real-world data and analysis has been presented. The proposed training pipeline has been tested for device heterogeneity. It has been found that even for an unknown test device, the localization accuracy is significantly good and 80% of the errors lies within 4m. [ABSTRACT FROM AUTHOR]

Detailed knowledge of a soil's lime requirement (LR) is a prerequisite for a demand-based lime fertilization to achieve the optimum soil pH and thus sustainably increasing soil fertility and crop yields. LR can be directly determined by the base neutralizing capacity (BNC) obtained by soil-base titration. For a site-specific soil acidity management, detailed information on the within-field variation of BNC is required. However, soil-base titrations for BNC determination are too laborious to be extensively applied in routine soil testing. In contrast, visible and near-infrared spectroscopy (visNIRS) is a time and cost-effective alternative that can analyze several soil characteristics within a single spectrum. VisNIRS was tested in the laboratory on 170 air-dried and sieved soil samples of nine agricultural fields of a quaternary landscape in North-east Germany predicting the soil's BNC and the corresponding lime requirement (LRBNC) at a target pH of 6.5. Seven spectral pre-processing methods were tested including a new technique based on normalized differences (ND). Furthermore, six multivariate regression methods were conducted including a new method combining a forward stagewise subset selection algorithm with PLSR (FS-PLSR). The models were validated using an independent sample set. The best regression model for most target variables was FS-PLSR combined with the second Savitzky-Golay derivation as pre-processing method achieving R2s from 0.68 to 0.82. Finally, the performance of the direct prediction of LRBNC (R2 = 0.68) was compared with an indirect prediction that was calculated by the predicted BNC parameters. This resulted in slightly higher correlation coefficients for the indirect method with R2 = 0.75. [ABSTRACT FROM AUTHOR]