Feature subset selection using association rule mining and JRip classifier

Feature selection is an important task in many fields such as statistics and machine learning. It aims at preprocessing step that include removal of irrelevant and redundant features and the retention of useful features. Selecting the relevant features increases the accuracy and decreases the computational cost. Feature selection also helps to understand the relevant data, addressing the complexity of dimensionality. In this paper, we have proposed a technique that uses JRip classifier and association rule mining to select the most relevant features from a data set. JRip extracts the rules from a data set and then association rules mining technique is applied to rank the features. Twenty datasets are tested ranging from binary class problem to multi-class problem. Extensive experimentation is carried out and the proposed technique is evaluated against the performance of various familiar classifiers. Experimental results demonstrate that while employing less number of features the proposed method achieves higher classification accuracy as well as generates less number of rules.   Key words: Feature subset selection, association rules mining, JRip, J48, Ridor, PART.