Reducing the Size of Feature Set by Using Modified MCA

Abstract

Reducing the size of a feature set, without altering the original representation, is an essential data processing step prior to applying a learning algorithm. The removal of irrelevant and redundant information improves the performance of machine learning algorithms. In this paper, Modified-Multiple Correspondence Analysis (Modified-MCA) is introduced. It integrates the correlation and reliability information between each feature and each class. Moreover, the proposed method contributes the optimal p-value to improve the reliability. To evaluate the performance of proposed method, experiments are carried out on ten benchmark datasets. In the experiments, three classifiers namely AdaBoost, Decision Table, JRip are used to verify that the output feature dataset produced by proposed method outperforms. Using three different classifiers is to get more accurate average classification results than using one classifier. The proposed Modified-MCA demonstrates reducing the size of the feature subspace and promising classification results. Moreover, the results performs that the propose method is better than other well-known feature selection methods; MCA, Information Gain and Relief.