A novel binary horse herd optimization algorithm for feature selection problem

Feature selection (FS) is an essential step for machine learning problems that can improve the performance of the classification by removing useless features from the data set. FS is an NP-hard problem, so meta-heuristic algorithms can be used to find good solutions for this problem. Horse herd Optimization Algorithm (HOA) is a new meta-heuristic approach inspired by horses 'herding behavior. In this paper, an improved version of the HOA algorithm called BHOA is proposed as a wrapper-based FS method. To convert continuous to discrete search space, S-Shaped and V-Shaped transfer functions are considered. Moreover, to control selection pressure, exploration, and exploitation capabilities, the Power Distance Sums Scaling approach is used to scale the fitness values of the population. The efficiency of the proposed method is estimated on 17 standard benchmark datasets. The implementation results prove the efficiency of the proposed method based on the V-shaped category of transfer functions compared to other transfer functions and other wrapper-based FS algorithms.

Automated summary

Feature selection is important for improving the performance of classification by removing useless features from the data set in machine learning problems. This paper proposes an improved version of Horse herd Optimization Algorithm (HOA) called BHOA as a wrapper-based feature selection method. S-Shaped and V-Shaped transfer functions are considered to convert continuous search space to discrete search space. Furthermore, the Power Distance Sums Scaling approach is used to control selection pressure, exploration, and exploitation capabilities. The implementation results on 17 standard benchmark datasets demonstrate the efficiency of the proposed method based on the V-shaped transfer functions compared to other transfer functions and other wrapper-based feature selection algorithms.