A hybridized approach for feature selection using Ant Colony Optimization and Ant-Miner for classification

This work presents an Ant Colony Optimization-based approach to feature selection that works in tandem with an ACO classifier (Ant-Miner) in a wrapper approach to improve the classification accuracy of the Ant-Miner with a small and appropriate feature subset. The objective is to analyze the performance of five ACO algorithms on the feature selection problem and the performance of the proposed FS-ACO/Ant-Miner system when compared to other feature selection for classification algorithms. The experimental results indicate that the hybridized approach performs comparatively well in discriminating input features and also achieves high classification accuracy especially for data sets with higher number of features.