Spotted Hyena Optimization Algorithm With Simulated Annealing for Feature Selection

• Published in: IEEE access Vol. 7; pp. 71943 - 71962
• Main Authors: Jia, Heming, Li, Jinduo, Song, Wenlong, Peng, Xiaoxu, Lang, Chunbo, Li, Yao
• Format: Journal Article
• Language: English
• Published: Piscataway IEEE 2019; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Algorithms; Biological system modeling; classification; Classification algorithms; Feature extraction; Feature selection; Genetic algorithms; Heuristic methods; Hybrid optimization; Iterative methods; Machine learning algorithms; Optimization; Optimization algorithms; SHO optimization; Simulated annealing; spotted hyena optimization algorithm
• DOI: 10.1109/ACCESS.2019.2919991
• ISSN: 2169-3536

The purpose of this paper is to propose a new hybrid metaheuristic to solve the problem of feature selection. Feature selection problem is the process of finding the most relevant subset based on some criteria. A hybrid metaheuristic is a new trend in the development of optimization algorithms. In this paper, two different hybrid models are designed based on spotted hyena optimization (SHO) for feature selection problem. The SHO algorithm can find the optimal or nearly optimal feature subset in the feature space to minimize the given fitness function. In the first model, the simulated annealing algorithm (SA) is embedded in the SHO algorithm (called SHOSA-1) to enhance the optimal solution found by the SHO algorithm after each iteration. In the second model, SA enhances the final solution obtained by the SHO algorithm (called SHOSA-2). The performance of these methods is evaluated in 20 datasets in the UCI repository. The experiments show that SHOSA-1 performs better than the native algorithm and SHOSA-2. And then, SHOSA-1 is compared with six state-of-the-art optimization algorithms. The experimental results confirm that SHOSA-1 improves the classification accuracy and reduces the number of selected features compared with other wrapper-based optimization algorithms. That proves the excellent performance of SHOSA-1 in spatial search and feature attribute selection.