Hybrid cuckoo search algorithm for simultaneous feature and classifier selection

Abstract

In literature, there are many supervised learning algorithms presented and applied in various problem domains. However, none of them could consistently perform well over all the datasets. This paper presents a novel approach for simultaneous selection of optimal feature subset and classifier for a given dataset. For large scale problems, this would require to search a huge solution space. Therefore, an efficient meta-heuristic known as cuckoo search (CS) algorithm has been utilized for searching; objective is to select an optimal combination of feature subset and classifier that minimizes the classification error rate, and reduces the dimensionality of feature vector. The proposed method (CSFCS) has been validated on several benchmark datasets. The results suggest that CS is suitable for the task resulting in higher classification accuracy with minimal feature subset and CSFCS is a generalized and practical approach.