An effective method for solving feature selection problems based on multi-strategy Hippo optimization algorithm

Abstract

Metaheuristic algorithms are important methods for solving practical optimization problems. Accurate, efficient, and stable metaheuristic algorithms have become the main goal of optimization research. Hippopotamus optimization algorithm (HO) is a new algorithm with good optimization effect proposed by three biological behaviors of hippopotamus: position change, defense, and avoidance of predators. However, due to the many restrictions of hippopotamus itself and the incomplete exploitation of hippopotamus’ physiological functions, in the exploration and exploitation stages, the parameter settings are single and limited in range, and the position change strategies under different conditions are less different, resulting in the algorithm’s limited search ability, tendency to converge prematurely and fall into the local optimal trap, poor exploration accuracy, and other shortcomings, making it impossible to obtain good results when facing optimization problems. In response to these shortcomings, this paper proposes a multi-strategy Hippopotamus optimization algorithm (MSHO), which achieves better performance by introducing three strategy improvement methods: adaptive perturbation optimization, autonomous optimal screening and new avoidance and defense. The effectiveness of MSHO is verified by comparing it with 10 excellent algorithms on the two test sets of CEC2017 and CEC2020. MSHO ranked first in the Friedman test results of the two experimental test sets, and in the two test sets, the optimization effect of MSHO was improved by 47.21% and 45.66%, respectively, compared with HO. Parameter perturbation analysis experiments were conducted, and the optimization effect of the improved strategy on the algorithm was obtained through ablation experiments. In addition, feature selection (FS) helps to improve the accuracy and efficiency of data analysis in fields such as engineering design and image analysis. In order to verify that MSHO has good effects in FS applications, it was compared with 10 excellent algorithms on 23 FS data sets. The Friedman test results of MSHO on 23 experimental FS data sets show that the optimization effect of MSHO is also better than other comparison algorithms, and the optimization effect is improved by 32.11% compared with HO, proving that MSHO performs better than the original HO and other comparison algorithms in the application of FS. MSHO and HO are tested for exploration and exploitation balance and population diversity, and it is finally concluded that MSHO has good stability and population diversity.