STUDY AND DEVELOPMENT OF NOVEL FEATURE SELECTION FRAMEWORK FOR EFFECTIVE DATA CLASSIFICATION

Abstract

Feature selection is the process of removing irrelevant features can be extremely useful in reducing the dimensionality of the data to be processed in reducing execution time and improving predictive accuracy of the classifier. Selecting the right set of features for classification is one of the most important problems in designing a good classifier. This is an important stage of pre-processing and is one two ways of avoiding the curse of dimensionality (the other is feature extraction). There are two approaches in Feature selection known as Forward selection and backward elimination. The first step is to create a set of n ranking lists ranking lists using corresponding rankers and the second is to select the combination function i.e. the function that will transform the ranking lists obtained in the first step into one single ranking list. The second step is the crucial step as it contains the combining method. We found that whereas different feature selectors like information gain, Gain Ratio, Chi-Square, PCA, applied to the datasets. We reviewed eight feature ranking techniques they are information gain, gain ratio, symmetrical uncertainty, relief F and one R attribute evaluation, Chi-square, and SVM attribute level, Filter Attribute level. We examined classification models that are built using various classification techniques such as naïve bayes, k-Star, random forest, support vector machine,J48,JRIP,Decision Table,Miltilayer Perception. The purpose of the proposed method is to reduce the computational complexity and increase the classification accuracy of the selected feature subsets. The dependence between two attributes is determined based on the probabilities of their joint values that contribute to positive and negative classification decisions. If there is an opposing set of attribute values that do not lead to opposing classification decisions, the two attributes are considered independent, otherwise dependent. One of them can be removed and thus the number of attributes is reduced.