A Comprehensive Study of Grey Wolf Optimizer Variants for Optimizing Feature Selection in High-Dimensional Data

Metaheuristic algorithms, notably the Grey Wolf Optimizer (GWO), are widely used for wrapper-based feature selection due to their superior ability to balance exploration and exploitation. We conduct a comparative evaluation of the original GWO and eight established variants on ten benchmark datasets from diverse domains. All nine selectors are tested under a unified wrapper protocol. The evaluation encompasses accuracy, composite fitness, feature-reduction rate, runtime, Fisher index, and selection consistency. Statistical significance is assessed using the Friedman test with Nemenyi post-hoc comparisons. Multi-Swarm GWO (MS-GWO) achieves the highest average classification accuracy on most datasets and the best overall optimization quality, ranking first with a mean rank of 2.55 across all metrics. Chaotic GWO (CGWO) delivers the strongest dimensionality reduction and the lowest average runtime, indicating superior computational efficiency. Adaptive GWO (AGWO) performs worst in accuracy and feature reduction, limiting its utility in high-dimensional settings. The original GWO offers balanced performance and serves as a stable baseline. On average, MS-GWO’s fitness is 23.21% lower than the pooled mean of the other variants, indicating superior optimization performance. These results guide the selection of GWO variants, with MS-GWO recommended for accuracy and overall optimization, and CGWO for dimensionality reduction and speed.