A Modified Variable Velocity Strategy Particle Swarm Optimization Algorithm for Multi-objective Feature Selection

With the ongoing advancement of big data and information technology, the efficient extraction of valuable feature information from vast existing datasets has become a fundamental task. The task is called feature selection, which is of paramount importance in contemporary data mining. It can eliminate irrelevant or redundant features and select the most relevant and useful features from the raw data to improve the model's generalization ability and accuracy. This process helps reduce modeling costs and shorten execution time. In this context, a multi-objective feature selection problem is proposed with the objectives of minimizing both the number of features and the classification error rate. To address this multi-objective problem more effectively, this paper designs a modified variable velocity strategy particle swarm optimization algorithm. The algorithm incorporates whale encircling and flipping, along with an inertia weight updating strategy for random perturbation, known as WETVVS-MOPSO. The results show that WETVVS-MOPSO significantly outperforms its competitors.