Modified-MCA Based Feature Selection Model for Preprocessing Step of Classification

Abstract

Feature subset selection is a technique for reducing the attribute space of a feature set. In other words, it is identifying a subset of features by removing irrelevant or redundant features. A good feature set that contains highly correlated features with the class improves not only the efficiency of the classification algorithms but also the classification accuracy. A novel metric that integrates the correlation and reliability information between each feature and each class obtained from multiple correspondence analysis (MCA) is currently the popular solution to score the features for feature selection. However, it has the disadvantage that p-value which examines the reliability is conventional confidence interval. In this paper, modified multiple correspondence analysis (M-MCA) is used to improve the reliability. The efficiency and effectiveness of proposed method is demonstrated through extensive comparisons with MCA using five benchmark datasets provided by WEKA and UCI repository. Naïve bayes, decision tree and jrip are used as the classifiers. The classification results, in terms of classification accuracy and size of feature subspace, show that the proposed Modified-MCA outperforms three other feature selection methods, MCA, information gain, and relief.

Exploring Hidden Coherent Feature Groups and Temporal Semantics for Multimedia Big Data Analysis

“…In the literature, many existing feature selection methods can be classified into two categories: univariate and multivariate [50]. Univariate methods, such as information gain and chi-square measure [50,54], consider the effect of each feature on a class separately without considering the inter-dependence among features.…”

“…To address this issue, research efforts have been directed towards various essential aspects like feature selection [48,49,50,21], training data selection [51,15], and classifier selection/fusion [52,53]. Among them, feature selection is considered especially applicable in big data analysis because it eliminates features with little predictive information, which also reduces the dimensionality of data and allows the learning algorithms to operate faster and more effectively [50]. In addition, research shows that a well designed feature selection method can not only handle high-dimensional data sets, but also successfully enhance classification performance in coping with imbalanced data [49,21].…”

“…Univariate methods, such as information gain and chi-square measure [50,54], consider the effect of each feature on a class separately without considering the inter-dependence among features. By contrast, multivariate methods, such as correlation-based feature selection [55], take features' interdependence into account.…”