Contribution Based Adaptive Multi-population Artificial Bee Colony Algorithm for Feature Interaction Selection on High-Dimensional Data

Feature interaction selection methods often encounter computational challenges due to the interactions between features in high-dimensional data. To tackle this challenge, we propose a contribution-based adaptive multi-population artificial bee colony algorithm (AMP-ABC) for feature interaction selection. AMP-ABC divides the original problem into several smaller sub-problems, with each sub-problem optimized by a corresponding population in a divide-and-conquer manner. A contribution-based feature grouping strategy is introduced, which groups features with large contributions into the same subspace, thereby increasing the likelihood of selecting feature interactions associated with the class label. During the search process, an onlooker bee stage that enhances exploration is incorporated to improve the algorithm’s search capability. Additionally, an adaptive iteration mechanism is employed to automatically adjust the number of iterations for each population. AMP-ABC is tested on a combination of artificially generated and real biological datasets to evaluate its performance. Experimental results on synthetic data demonstrate that AMP-ABC outperforms existing state-of-the-art evolutionary computation-based feature selection methods. Furthermore, empirical results on actual biological data confirm the performance of the proposed method.