Information Gain Measured Feature Selection to Reduce High Dimensional Data

Abstract

While demand of the massive amount of data to be more effective and efficient mining strategies is increasing significantly, practitioners and researchers are trying to develop scalable machine learning algorithms and strategies in turning mountains of data into nuggets. High dimension of data makes the memory, storage requirements and computational costs increased significantly. Therefore, reducing dimension can mainly improve learning performance. Feature selection, a data preprocessing technique, is effective and efficient to enhance data mining, data analytics and machine learning. Most feature selection algorithms have been trying to eliminate irrelevant features. However, removing only irrelevant features is not enough to get the best insight and patterns. Not only irrelevant features but also redundant features can degrade learning performance. Feature selection methods which can eliminate both irrelevant and redundant features are demanding in high dimensional data analytics. To solve this problem, information gain measured feature selection is presented in this work.