Boosting feature selection efficiency with IMVO: Integrating MVO and mutation-based local search algorithms

Feature selection is crucial in machine learning and data mining, significantly impacting model performance and efficiency by reducing dimensionality, mitigating overfitting, and improving interpretability. Effective feature selection lowers computational costs, making it indispensable for processing large datasets. In this research, we introduce the Improved Multi-Verse Optimizer (IMVO) algorithm, a novel feature selection method that integrates the Multi-Verse Optimizer (MVO) with local search algorithms (LSAs). This hybrid approach leverages MVO's global search abilities to explore solution spaces and identify promising feature subsets while using LSAs for fine-tuning to achieve optimal solutions. Our method balances exploration and exploitation, focusing resources on the most promising regions, thus enhancing feature selection effectiveness and reducing computational costs. Key contributions include the development of a novel LSA based on mutation and random selection, improving exploitation and population diversity within IMVO, and the introduction of a two-phase LSA mechanism to avoid local optima and increase solution diversity. The exceptional performance of the proposed IMVO algorithm is thoroughly evaluated using 23 benchmark functions. Extensive experiments on 20 UCI datasets demonstrate that IMVO surpasses five well-known optimization algorithms in terms of accuracy, feature selection, and fitness value. Statistical analysis using the Wilcoxon signed-rank test confirms the significance of these improvements, underlining IMVO's robustness and reliability. Although incorporating LSAs increases computational complexity, the resulting performance enhancements justify this trade-off. Our findings establish IMVO as a powerful tool for feature selection and classification, with potential applications across various optimization scenarios.