Improved salp swarm optimization algorithm based on a robust search strategy and a novel local search algorithm for feature selection problems

Abstract

The enormous challenge in data science and data mining for knowledge extraction is confronting an expansive high number of data dimensions. Because the process of extracting knowledge from data can become more complex and memory-consuming. Not only the presence of all features doesn't help the learning process, but also it can sometimes decrease the model's efficiency. To enhance the model's efficiency and reduce the problem's complexity, various feature selection algorithms are designed and implemented. In this paper, a novel and highly effective algorithm based on the salp swarm optimization algorithm for solving complex problems is proposed for feature selection. In the proposed method, an unexpected event that causes the chain to break apart (such as hitting an obstacle or the death of the chain leader, etc.) is modeled which is not taken into account in the salp swarm optimization algorithm. Also, the exploration capability is improved by modifying the updating the position of the chain leader. Additionally, an innovative local search algorithm has been embedded into the proposed algorithm to enhance its exploitation. The proposed approach is implemented on 14 datasets, and the results are compared by two terms, classification accuracy and number of selected features. Additionally, the effectiveness of the proposed method is tested on 2 widely used chemical datasets. The modifications that are applied to the standard salp swarm algorithm reduce the probability of getting stuck in the local optimum and simultaneously, increase the diversity of solutions. The results show that the proposed algorithm has performed significantly better than other algorithms in solving the feature selection problem.