Performance Analysis of Filter Based Feature Selection Techniques in Tree Based Classification Methods

Abstract

Classification has been called the most influential development in Data Mining and Machine Learning in the past decade. The idea of classification is to find the class of the unknown objects based on their attributes. In this thesis, the performance of decision tree based classification methods is analysed with feature selection methods; Chi-square and Relief. The feature selection process chooses optimal subset of features according to objective function. These feature selection method helps to remove unnecessary attributes from the high dimensional dataset, thus improves the efficiency of the classification algorithms. The performance of feature section methods; Chisquare and Relief

were compared in Tree based classification methods; C4.5, CART, LMT and Random Tree. The study shows that the Chi-square feature selection method is more suitable while using with LMT followed by C4.5, Random Tree and CART respectively. In case Relief based feature selection method LMT gives the best result followed by CART, C4.5 and Random Tree respectively.