Boosting Multiverse Optimizer by Simulated Annealing for Dimensionality Reduction

Abstract

Background: Because of The Multi-Verse Optimizer (MVO) has gained popularity in feature selection due to its strong global and local search capabilities. However, its effectiveness diminishes when tackling high-dimensional datasets due to the exponential growth of the search space and a tendency for premature convergence. Objective: This study aims to enhance MVO’s performance by integrating it with the Simulated Annealing Algorithm (SAA), creating a hybrid model that improves search convergence and optimizes feature selection efficiency. Methods: A High-level Relay Hybrid (HRH) architecture is proposed, where MVO identifies promising regions of the feature space and passes them to SAA for local refinement. The resulting MVOSA-FS model was evaluated on ten high-dimensional benchmark datasets from the Arizona State University (ASU) repository. Support Vector Machine (SVM) classifiers were used to assess the classification accuracy. MVOSA-FS achieved superior performance compared to six state-of-the-art feature selection algorithms: Atom Search Optimization (ASO), Equilibrium Optimizer (EO), Emperor Penguin Optimizer (EPO), Monarch Butterfly Optimization (MBO), Satin Bowerbird Optimizer (SBO), and Sine Cosine Algorithm (SCA). Results: The proposed model yielded the lowest average classification error rate (1.45%), smallest standard deviation (0.008), and most compact feature subset (0.91%). The hybrid MVOSA-FS model effectively balances exploration and exploitation, delivering robust and scalable performance in feature selection for high-dimensional data. Conclusion: This hybridization approach demonstrates improved classification accuracy and reduced computational burden.