Ant colony optimization and mutual information hybrid algorithms for feature subset selection in equipment fault diagnosis

This paper presents a method to determine optimum feature subset selection with ant colony optimization and mutual information hybrid algorithms. We present details of the algorithm, design and implementation of feature subset selection using ant colony algorithms. The best compound features found by ant colony algorithms are verified by multiple regression models and are used to construct fault prediction models. A case study of machinery tool wear-out prediction is presented. The fairly good agreement between the prediction result and real tool wear-out data demonstrates the viability of the feature subset selection method for diagnosis applications.