A Novel Binary Grey Wolf Optimizer Algorithm With a New Dynamic Position Update Mechanism for Feature Selection Problem: A Novel Binary Grey Wolf Optimizer Algorithm With a New Dynamic Position…: F. Erdoğan Et Al.

Abstract

Feature selection (FS) is one of the basic preprocessing steps in data mining and is a challenging binary optimization problem. FS is the process of determining the subset that can best represent the dataset by removing features that have little impact from a given dataset without affecting performance and accuracy. In this paper, Binary Dynamic Grey Wolf Optimization Algorithm (binDGWO) is proposed for the solution of binary optimization problems. To binaryize the Grey Wolf Optimization Algorithm (GWO), the original position update equation was binaryized using the logical XOR operator to achieve a balance between local search and global search. In addition, a simple and effective innovation has been introduced to the position update equation with the dynamic coefficient method. This method has been developed to make the affected solution better by determining and applying the effects of the wolves in the lead team on the position according to the solution quality. The performance of binDGWO over FS is compared to the performance of over 20 different algorithms, including binary variants of GWO and different binary metaheuristics. 41 datasets with different numbers of samples and features were used in the experiments. Various performance metrics were used to determine the superiority of the methods over each other. In addition, the Friedman test was performed to statistically evaluate the results of the methods. According to the performance metrics and the Friedman test results, it was seen that the proposed algorithm has better results than other binary variants of GWO, and when the comparison results with other metaheuristic algorithms are examined, it is generally more successful and effective. In conclusion, it can be said that binDGWO is a simple, effective, and efficient binary method and it achieves its purpose. © The Author(s), under exclusive licence to Springer-Verlag GmbH Germany, part of Springer Nature 2024.