A New Hybrid PSO-HHO Wrapper Based Optimization for Feature Selection

Datasets used in data analysis often contain irrelevant or redundant attributes. These attributes hinder the performance of predictive models. Therefore, an effective preprocessing feature selection procedure is essential to identify the relevant features and eliminate unnecessary ones. Metaheuristic algorithms, inspired mainly by nature, are strong candidates for the feature selection process, as they can efficiently search large solution spaces. Metaheuristic algorithms provide flexible strategies for complex optimization problems such as traveling salesman problem, scheduling problems, nonlinear integer programming, and multi-objective optimization. Reasonably balancing exploration and exploitation will increase the search algorithm’s performance. This research combines Particle Swarm Optimization (PSO) and Harris Hawk Optimization (HHO) to form a Hybrid PSO-HHO approach to enhance the feature selection process. The wrapper-based approach employs the K-nearest neighbors classifier using the Euclidean distance metric to identify the optimal solutions. The performance of the proposed binary algorithm is evaluated using four standard benchmark datasets obtained from the UCI repository. Classification accuracy and the number of selected features are used as performance indicators. Experimental results demonstrated that the hybrid model achieved mean accuracies of 95.23%, 97.61%, and 92.24% across various datasets, outperforming traditional PSO and HHO approaches.