Correlation-based Feature Subset Research Articles

ECAmyloid: An amyloid predictor based on ensemble learning and comprehensive sequence-derived features

Amyloid fibrils formed by the mis-aggregation of amyloid proteins can lead to neuronal degenerations in the Alzheimer's disease. Predicting amyloid proteins not only contributes to understanding physicochemical properties and formation mechanism of amyloid proteins, but also has significant implications in the amyloid disease treatment and the development of a new purpose for amyloid materials. In this study, an ensemble learning model with sequence-derived features, ECAmyloid, is proposed to identify amyloids. The sequence-derived features including Pseudo Position Specificity Score Matrix (Pse-PSSM), Split Amino Acid Composition (SAAC), Solvent Accessibility (SA), and Secondary Structure Information (SSI) are employed to incorporate sequence composition, evolutionary and structural information. The individual learners of the ensemble learning model are selected by an increment classifier selection strategy. The final prediction results are determined by voting of prediction results of multiple individual learners. In view of the imbalanced benchmark dataset, the Synthetic Minority Over-sampling Technique (SMOTE) is adopted to generate positive samples. To eliminate irrelevant features and redundant features, correlation-based feature subset (CFS) selection combined with a heuristic search strategy is performed to obtain the optimal feature subset. Experimental results indicate that the ensemble classifier achieves an accuracy of 98.29%, a sensitivity of 0.992, a specificity of 0.974 on the training dataset using the 10-fold cross validation, far higher than the results obtained by its individual learners. Compared with the original feature set, the accuracy, sensitivity, specificity, MCC, F1-score, G-Mean of the ensemble method trained by the optimal feature subset are improved by 1.05%, 0.012, 0.01, 0.021, 0.011 and 0.011, respectively. Moreover, the comparison results with existing methods on two same independent test datasets demonstrate that the proposed method is an effective and promising predictor for large-scale determination of amyloid proteins. The data and code used to develop ECAmyloid has been shared to Github, and can be freely downloaded at https://github.com/KOALA-L/ECAmyloid.git.

Potential of ATR-FTIR-Chemometrics in Covid-19: Disease Recognition.

The COVID-19 pandemic has caused major disturbances to human health and economy on a global scale. Although vaccination campaigns and important advances in treatments have been developed, an early diagnosis is still crucial. While PCR is the golden standard for diagnosing SARS-CoV-2 infection, rapid and low-cost techniques such as ATR-FTIR followed by multivariate analyses, where dimensions are reduced for obtaining valuable information from highly complex data sets, have been investigated. Most dimensionality reduction techniques attempt to discriminate and create new combinations of attributes prior to the classification stage; thus, the user needs to optimize a wealth of parameters before reaching reliable and valid outcomes. In this work, we developed a method for evaluating SARS-CoV-2 infection and COVID-19 disease severity on infrared spectra of sera, based on a rather simple feature selection technique (correlation-based feature subset selection). Dengue infection was also evaluated for assessing whether selectivity toward a different virus was possible with the same algorithm, although independent models were built for both viruses. High sensitivity (94.55%) and high specificity (98.44%) were obtained for assessing SARS-CoV-2 infection with our model; for severe COVID-19 disease classification, sensitivity is 70.97% and specificity is 94.95%; for mild disease classification, sensitivity is 33.33% and specificity is 94.64%; and for dengue infection assessment, sensitivity is 84.27% and specificity is 94.64%.

Predicting Hub Genes of Glioblastomas Based on a Support Vector Machine Combined with CFS Algorithms

Aim: To search the genes related to the mechanisms of the occurrence of glioma and to try to build a prediction model for glioblastomas. Background: The morbidity and mortality of glioblastomas are very high, which seriously endanger human health. At present, the goals of many investigations on gliomas are mainly to understand the cause and mechanism of these tumors at the molecular level and to explore clinical diagnosis and treatment methods. However, there is no effective early diagnosis method for this disease, and there are no effective prevention, diagnosis, or treatment measures. Methods: Firstly, the gene expression profiles derived from GEO were downloaded. Then, differentially expressed genes (DEGs) in the disease samples and the control samples were identified. After that, GO and KEGG enrichment analyses of DEGs were performed by DAVID. Furthermore, the correlation- based feature subset (CFS) method was applied to the selection of key DEGs. In addition, the classification model between the glioblastoma samples and the controls was built by a Support Vector Machine (SVM) based on selected key genes. Results and Discussion: Thirty-six DEGs, including 17 upregulated and 19 downregulated genes, were selected as the feature genes to build the classification model between the glioma samples and the control samples by the CFS method. The accuracy of the classification model by using a 10-fold crossvalidation test and the independent set test was 76.25% and 70.3%, respectively. In addition, PPP2R2B and CYBB can also be found in the top 5 hub genes screened by the protein-protein interaction (PPI) network. Conclusions: This study indicated that the CFS method is a useful tool to identify key genes in glioblastomas. In addition, we also predicted that genes such as PPP2R2B and CYBB might be potential biomarkers for the diagnosis of glioblastomas.

Intensity prediction of tropical cyclone using multilayer multi-block local binary pattern and tree-based classifiers over North Indian Ocean

Local binary pattern (LBP) is an extensively used method in image analysis and pattern recognition related works across various domains. In this paper, we explore cyclone cloud evolution patterns using a modified LBP. The multilayer multi-block local binary pattern (MMLBP) is an extended version of Completed LBP (CLBP) with multiple blocks and many layers. A tropical cyclone (TC) image is converted into 3 × 3 identical blocks to generate central pixels using CLBP magnitude descriptor block-wise (CLBP_MB). These central pixels are used to find the next layer, which is further divided into 3 × 3 blocks to generate central pixels by CLBP_MB descriptor. This technique will iterate until the dimension of the final layer reached to 1 × 1. These central pixels are extracted from each layer and gather in a vector called a feature vector. The proposed method is applied to 600 tropical cyclone (TC) images and each image generates one feature vector. Hence, 600 vectors are passed through tree-based classifiers to classify cyclone images of various classes. The proposed method of feature extraction and classification reaches a maximum of 84.66% accuracy using Random Forest (RF) classifier. The root mean squared error (RMSE) of the MMLBP method is 9.27 kt, which is better than the LBP and MBLBP approaches. Finally, the size of the original feature vector is reduced to 97.4% using correlation based feature subset selection method.

Quality grade classification of China commercial moxa floss using electronic nose: A supervised learning approach.

Moxa floss is the primary material used in moxibustion, an important traditional Chinese medicine therapy that uses ignited moxa floss to apply heat to the body for disease treatment. Till date, there is no available data regarding quality control of different grades of moxa floss. The objectives of this study were to explore the probative value of the electronic nose (e-nose) in differentiating different quality grades of commercial moxa floss sold in China, and to investigate if data mining techniques could be used to optimize the sensor array while retaining classification accuracy of the samples. The e-nose with 12 metal oxide semiconductor type sensors was used to analyze the odor profiles of 15 commercial moxa floss samples of different quality grades. Feature selection algorithms using principal component analysis (PCA) and BestFirst (BC) coupled with correlation-based feature subset selection (CfsSubsetEval) method were used to obtain the most efficient feature subsets. Results for the BC feature selection method identified 3 optimized sensors (S2, S6, and S11), suggesting that aromatic compounds relate more to the identification of the samples. Radial basis function (RBF), multilayer perceptron (MLP), and random forests (RF) performed well in discriminating the samples, retaining prediction accuracies above 85%, which achieved cost-effectiveness and operational simplicity, while retaining prediction accuracy. The e-nose could be a rapid and nondestructive method for objective preliminary classification of quality grades of moxa floss and may be used for future studies related to moxa products safety and quality.

Comparative Analysis of Feature Selection Algorithms for Computational Personality Prediction From Social Media

With the rapid growth of social media, users are getting involved in virtual socialism, generating a huge volume of textual and image contents. Considering the contents such as status updates/tweets and shared posts/retweets, liking other posts is reflecting the online behavior of the users. Predicting personality of a user from these digital footprints has become a computationally challenging problem. In a profile-based approach, utilizing the user-generated textual contents could be useful to reflect the personality in social media. Using huge number of features of different categories, such as traditional linguistic features (character-level, word-level, structural, and so on), psycholinguistic features (emotional affects, perceptions, social relationships, and so on) or social network features (network size, betweenness, and so on) could be useful to predict personality traits from social media. According to a widely popular personality model, namely, big-five-factor model (BFFM), the five factors are openness-to-experience, conscientiousness, extraversion, agreeableness, and neuroticism. Predicting personality is redefined as predicting each of these traits separately from the extracted features. Traditionally, it takes huge number of features to get better accuracy on any prediction task although applying feature selection algorithms may improve the performance of the model. In this article, we have compared the performance of five feature selection algorithms, namely the Pearson correlation coefficient (PCC), correlation-based feature subset (CFS), information gain (IG), symmetric uncertainly (SU) evaluator, and chi-squared (CHI) method. The performance is evaluated using the classic metrics, namely, precision, recall, f-measure, and accuracy as evaluation matrices.

Real-time anomaly detection framework using a support vector regression for the safety monitoring of commercial aircraft

The development of an automated health monitoring framework is critical for aviation system safety, especially considering the expected increase in air traffic over the next decade. Conventional approaches such as model-based and exceedance methods have a low detection accuracy and are limited to specific applications. This paper proposes a robust real-time health monitoring framework for detecting performance anomalies, which may impact system safety during flight operations, with high accuracy and generalized applicability. The proposed monitoring framework utilizes sensor data from commercial flight data recorders to predict possible flight performance anomalies. Decimation, a signal processing technique, in conjunction with Savitzky-Golay filtering is utilized to preprocess the dataset and mitigate sampling rate and noise issues that prevent direct usage of historical flight data. Correlation-based feature subset selection is subsequently performed, and these features are used to train a support vector machine that predicts flight performance. With this model, performance anomalies in the test data are automatically detected based on deviations from the predicted flight behavior. The proposed monitoring framework was demonstrated to detect performance anomalies in real-time and exhibited accurate detection capabilities with high computational efficiency.

Prediction of ARDS Syndrome and CAD Using Multilayer Perceptron

Acute Respiratory Distress Syndrome (ARDS) and Coronary artery heart Disease (CAD) isa critical condition occurring in ill patients. Our proposed system is to predict ARDS and CAD in hospitalized patients using only physiological signals as heart rate and breathing rate. This based on hypothesis testing is developed to detect whether subjects signals deviate from their initial states. The approach is applied on mechanically ventilated subjects in the MIMIC II database. Hybrid method for CAD diagnosis, including risk factor identification using correlation based feature subset selection with particle swam optimization search method and K-Means clustering algorithms. This proposed system is increasing the efficiency and accuracy of predicting the ARDS and CAD diseases.

Text Classification of Cornell Movie Data using Data Mining with Feature Selection

Text Classification is branch of text mining through which we can analyze the sentiment of the movie data. In this research paper we have applied different preprocessing techniques to reduce the features from cornell movie data set. We have also applied the Correlation-based feature subset selection and chi-square feature selection technique for gathering most valuable words of each category in text mining processes. The new cornell movie data set formed after applying the preprocessing steps and feature selection techniques. We have classified the cornell movie data as positive or negative using various classifiers like Support Vector Machine (SVM), Multilayer Perceptron (MLP), Naive Bayes (NB), Bays Net (BN) and Random Forest (RF) classifier. We have also compared the classification accuracy among classifiers and achieved better accuracy i. e. 87% in case of SVM classifier with reduced number of features. The suggested classifier can be useful in opinion of movie review, analysis of any blog and documents etc.

Prediction and Analysis of Hub Genes in Renal Cell Carcinoma based on CFS Gene Selection Method Combined with Adaboost Algorithm.

Renal cell carcinoma (RCC) is the most common malignant tumor of the adult kidney. The aim of this study was to identify key genes signatures during RCC and uncover their potential mechanisms. Firstly, the gene expression profiles of GSE53757 which contained 144 samples, including 72 kidney cancer samples and 72 controls, were downloaded from the GEO database. And then differentially expressed genes (DEGs) between the kidney cancer samples and the controls were identified. After that, GO and KEGG enrichment analyses of DEGs were performed by DAVID. Furthermore, the correlation-based feature subset (CFS) method was applied to the selection of key genes of DEGs. In addition, the classification model between the kidney cancer samples and the controls was built by Adaboost based on the selected key genes. 213 DEGs including 80 up-regulated and 133 down-regulated genes were selected as the feature genes to build the classification model between the kidney cancer samples and the controls by CFS method. The accuracy of the classification model by using 5-folds cross-validation test and independent set test is 84.4% and 83.3%, respectively. Besides, TYROBP, CD4163, CAV1, CXCL9, CXCL11 and CXCL13 also can be found in the top 20 hub genes screened by proteinprotein interaction (PPI) network. It indicated that CFS is a useful tool to identify key genes in kidney cancer. Besides, we also predicted genes such as TYROBP, CD4163, CAV1, CXCL9, CXCL11 and CXCL13 that might target genes to diagnose the kidney cancer.

Comprehensive preference learning and feature validity for designing energy-efficient residential buildings using machine learning paradigms

Having a reliable approximation of heating load (HL) and cooling load (CL) is a substantial task for evaluating the energy performance of buildings (EPB). Also, the appearance of soft computing techniques has made many traditional methods antiquated. Thus, the main effort of this study was to evaluate the capability of several learning methods for appraising the HL and CL of a residential building. To this end, a proper dataset consisting of eight influential factors was provided. To simplify the problem, we executed feature validity by using a correlation-based feature subset selection (CfsSubsetEval) technique. The results of this process showed that wall area, overall height, orientation and glazing area have the most significant impact on the HL and CL simulation. After preparing the suitable dataset, sixteen learning methods namely, elastic net (EN), Gaussian process regression (GPR), least median of squares regression (LMSR), multiple linear regression (MLR), multi-layer perceptron regression (MPR), multi-layer perceptron (MLP), radial basis function regression (RBFR), sequential minimal optimization regression (SMOR), functions XNV, lazy K-star, lazy LWL, rules decision table (RDT), M5Rules, alternating model tree (AMT), directional path consistency (DPC), and Random Forest (RF) were developed in Weka environment to forecast the HL and CL variables. Referring to the results, it was concluded that RF, lazy K-star, RDT and AMT outperform other predictive models. Also, comparing the results with the results of the previous studies showed that the applied feature reduction not only did not disturb the learning process but also has enhanced the performance of models. Also, due to the excellent accuracy of the MLP, a formula was derived from the optimized structure of it to predict the HL and CL variables.

A Fuzzy Approach to Text Classification With Two-Stage Training for Ambiguous Instances

Sentiment analysis is a very popular application area of text mining and machine learning. The popular methods include support vector machine, naive bayes, decision trees, and deep neural networks. However, these methods generally belong to discriminative learning, which aims to distinguish one class from others with a clear-cut outcome, under the presence of ground truth. In the context of text classification, instances are naturally fuzzy (can be multilabeled in some application areas) and thus are not considered clear-cut, especially given the fact that labels assigned to sentiment in text represent an agreed level of subjective opinion for multiple human annotators rather than indisputable ground truth. This has motivated researchers to develop fuzzy methods, which typically train classifiers through generative learning, i.e., a fuzzy classifier is used to measure the degree to which an instance belongs to each class. Traditional fuzzy methods typically involve generation of a single fuzzy classifier and employ a fixed rule of defuzzification outputting the class with the maximum membership degree. The use of a single fuzzy classifier with the above-fixed rule of defuzzification is likely to get the classifier encountering the text ambiguity situation on sentiment data, i.e., an instance may obtain equal membership degrees to both the positive and negative classes. In this paper, we focus on cyberhate classification, since the spread of hate speech via social media can have disruptive impacts on social cohesion and lead to regional and community tensions. Automatic detection of cyberhate has thus become a priority research area. In particular, we propose a modified fuzzy approach with two-stage training for dealing with text ambiguity and classifying four types of hate speech, namely, religion, race, disability, and sexual orientation—and compare its performance with those popular methods as well as some existing fuzzy approaches, while the features are prepared through the bag-of-words and word embedding feature extraction methods alongside the correlation-based feature subset selection method. The experimental results show that the proposed fuzzy method outperforms the other methods in most cases.

Predictive Models for Classifying the Outcomes of Violence: Case Study for Thailand’s Deep South

Violence is now widely recognized as a public health problem because of its significant consequences on the health and wellness of people and it remains a growing problem in many countries including Thailand. Elucidating the factors related to violence can provide information that can help to prevent violence and decrease the number of injuries. This study explored predictive data mining models which have high interpretability and prediction accuracy in classifying the outcomes of violence. After data preprocessing, a set of 21,424 incidents occurring from 2004 to 2016 were obtained from the Deep South Coordination Centre database. A correlation-based feature subset selection and decision tree technique with embedded feature selection were used for variable selection and four data mining techniques were applied to classify the violent outcomes into physical injury and no physical injury. The findings revealed that regardless of the variable selection method, gun was selected as a risk factor of physical injury. Moreover, a decision tree model with three variables, gun, zone, and solid/sharp weapon outperformed a naive Bayes model in terms of accurate performance and interpretability. Decision tree and artificial neural network models have similar levels of performance in classifying the outcome of violence but in practical terms, a decision tree model is more interpretable than an artificial neural network model.

Preventing crimes ahead of time by predicting crime propensity in released prisoners using data mining techniques

Criminologists and psychologists around the world are finding new initiatives to identify criminals and understand crime scenes. This work focuses on predicting the occurrence of crimes for a released prisoner, based on crime propensity prediction, using a supervised machine learning technique. This original research is intended to design and develop a new dataset of 30 attributes that exists nowhere and is exclusively created to define prisoners so as to differentiate them by their propensity to crime using psychological and behavioural factors obtained from jails and assorted sources. The research incorporates an analysis of seven search methods, in tandem with seven subset evaluation techniques, to undertake feature selection, and nine classification algorithms for the classification of prisoners. It is found that the wolf search algorithm, used with the correlation-based feature subset evaluation technique and radial basis function classifier, performs best providing 97.8% precision, 97.5% recall and low error values.

Preventing crimes ahead of time by predicting crime propensity in released prisoners using Data Mining techniques

Criminologists and psychologists around the world are finding new initiatives to identify criminals and understand crime scenes. This work focuses on predicting the occurrence of crimes for a released prisoner, based on crime propensity prediction, using a supervised machine learning technique. This original research is intended to design and develop a new dataset of 30 attributes that exists nowhere and is exclusively created to define prisoners so as to differentiate them by their propensity to crime using psychological and behavioural factors obtained from jails and assorted sources. The research incorporates an analysis of seven search methods, in tandem with seven subset evaluation techniques, to undertake feature selection, and nine classification algorithms for the classification of prisoners. It is found that the wolf search algorithm, used with the correlation-based feature subset evaluation technique and radial basis function classifier, performs best providing 97.8% precision, 97.5% recall and low error values.

Selection of Neural Oscillatory Features for Human Stress Classification with Single Channel EEG Headset.

A study on classification of psychological stress in humans using electroencephalography (EEG) is presented. The stress is classified using a correlation-based feature subset selection method that efficiently reduces the feature vector length. In this study, twenty-eight participants are involved by filling in the perceived stress scale-10 (PSS-10) questionnaire and their EEG is also recorded in closed eye condition to measure the baseline stress. The recorded data is labelled on the basis of the stress level that is indicated by the participant's PSS score. The feature selection method has shown that, among the EEG oscillations, low beta, high beta, and low gamma are the most significant neural oscillations for classifying human stress. The proposed method not only reduces the time to build a classification model but also improves the classification accuracy up to 78.57% using a single channel wearable EEG device.

Comparison of Machine Learning Approaches with a General Linear Model To Predict Personal Exposure to Benzene.

Machine learning techniques (MLTs) offer great power in analyzing complex data sets and have not previously been applied to non-occupational pollutant exposure. MLT models that can predict personal exposure to benzene have been developed and compared with a standard model using a linear regression approach (GLM). The models were tested against independent data sets obtained from three personal exposure measurement campaigns. A correlation-based feature subset (CFS) selection algorithm identified a reduced attribute set, with common attributes grouped under the use of paints in homes, upholstery materials, space heating, and environmental tobacco smoke as the attributes suitable to predict the personal exposure to benzene. Personal exposure was categorized as low, medium, and high, and for big data sets, both the GLM and MLTs show high variability in performance to correctly classify greater than 90 percentile concentrations, but the MLT models have a higher score when accounting for divergence of incorrectly classified cases. Overall, the MLTs perform at least as well as the GLM and avoid the need to input microenvironment concentrations.

Classification of Breast Masses Using a Computer-Aided Diagnosis Scheme of Contrast Enhanced Digital Mammograms.

Contrast-enhanced digital mammography (CEDM) is a promising imaging modality in breast cancer diagnosis. This study aims to investigate how to optimally develop a computer-aided diagnosis (CAD) scheme of CEDM images to classify breast masses. A CEDM dataset of 111 patients was assembled, which includes 33 benign and 78 malignant cases. Each CEDM includes two types of images namely, low energy (LE) and dual-energy subtracted (DES) images. A CAD scheme was applied to segment mass regions depicting on LE and DES images separately. Optimal segmentation results generated from DES images were also mapped to LE images or vice versa. After computing image features, multilayer perceptron based machine learning classifiers that integrate with a correlation-based feature subset evaluator and leave-one-case-out cross-validation method were built to classify mass regions. When applying CAD to DES and LE images with original segmentation, areas under ROC curves (AUC) were 0.759 ± 0.053 and 0.753 ± 0.047, respectively. After mapping the mass regions optimally segmented on DES images to LE images, AUC significantly increased to 0.848 ± 0.038 (p < 0.01). Study demonstrated that DES images eliminated overlapping effect of dense breast tissue, which helps improve mass segmentation accuracy. The study demonstrated that applying a novel approach to optimally map mass region segmented from DES images to LE images enabled CAD to yield significantly improved performance.

The Impact of Feature Selection on Urban Land Cover Classification

Many of the studies in the literature about land cover classification are focused on the feature extraction and classification rather than feature selection. In this paper, the impact of feature selection on urban land cover classification is extensively analyzed. Three types of features namely spectral, texture, and size/shape features are used for this analysis. This analysis is carried out using three variations of a filter based feature selection method and three widely-known classification algorithms. The feature selection method used for the comparison is a multivariate filter method namely correlation-based feature subset selection where a feature subset evaluator and a search method are integrated. Best first search, genetic search, and greedy stepwise search are three different search methods used for this integration. The classification algorithms employed are Bayesian network, random forest, and support vector machine. The experimental results explicitly indicate that feature selection improves classification accuracy in all cases. Besides, according to the experimental results, random forest classifier is the most successful one among these three classifiers while both feature selection is applied and not applied. Largest improvement in the classification performance is obtained when greedy stepwise search based feature selection method and support vector machine classifier is applied together. Also, the contribution of spectral features to the performance of classification is more than size/shape and texture features.

Correlation-based feature subset selection technique for web spam classification

In past years, different machine learning algorithms and web spam features have been created to recognise the spam. The key part of progression of machine learning (ML) depends on the features being utilised. If we have features which correlate with each other then it is easy for ML to learn and if we have features which are very complex then ML may not be able to learn. It is the most imperative and basic area where the majority of the applications in a machine learning are going on. In this paper, correlation-based feature selection (CFS) technique (with best-first search) is used which selects features that are most efficient. Two datasets (WebSpam-UK2006 and WebSpam-UK2007) and four classifiers (Naive Bayes, J48, random forest and AdaBoost) are used for conducting the experiment. The results have shown significant improvement in AUC (area under receiver operating characteristic curve) for Naive Bayes and J48.