Feature Selection Algorithm Based on Association Rules

Abstract

Feature selection is one of the key issues in pattern recognition. The quality of the feature selection has a direct impact on the classification accuracy and generalization performance of the classifier. In order to reduce the size of the feature subset and improve the efficiency of the algorithm without reducing the accuracy, this paper proposes a feature selection algorithm based on association rules, ARFS. The algorithm uses association rules to mine the frequent 2-items set of the feature attributes and category in the dataset. Then the algorithm sorts the features according to the confidence of the frequent 2-items set, and then combines the sequential forward selection method, and uses the classification performance of the decision tree classifier as the evaluation criteria of the feature subsets. In terms of experiments, this paper selects five public datasets in the UCI Machine Learning Repository to conducted three comparison experiments of feature subset selection, learning algorithm accuracy and runtime for four feature selection algorithms. Experimental results show that the ARFS is superior to the contrast method in the feature subset size and the accuracy. However, the ARFS is slightly inferior to the ReliefF algorithm in the runtime.