Optimizing feature selection through a hybrid bi-objective genetic algorithm for enhanced classification

With the growing volume of data, efficient feature selection (FS) has become essential for enhancing classification models. FS aims to streamline models by identifying and removing irrelevant, redundant, or noisy features, potentially boosting predictive accuracy. This paper introduces the Hybrid Bi-Objective Genetic Algorithm for Feature Selection Problem (HBiOGA-FSP), which addresses key limitations of traditional FS methods, such as handling noisy or correlated features and exploring complex feature combinations. The algorithm integrates advanced mathematical models, combining tabu search and hill-climbing techniques within its local search process to improve optimization. It also incorporates a mutation operator based on a dynamic distribution Lévy flight, enhancing adaptability and efficiency in finding global optimal solutions. As a wrapper FS model, it uses the C4.5 classifier to evaluate feature subsets. Experimental evaluation on 20 real-world datasets demonstrated an average accuracy improvement of 11.9% and a 42.6% reduction in false alarms, highlighting its superior efficiency and robustness compared to seven state-of-the-art algorithms. These results confirm the potential of HBiOGA-FSP in supporting decision-making across various fields.