A Multi-Objective Hybridized Metaheuristic Optimization Technique for Discriminative Feature Selection from High-Dimensional Data

Feature selection is considered a fundamental preprocessing step in machine learning and data mining research area, specifically for high-dimensional datasets that includes the redundant and irrelevant features that affects the classification performance and computational efficiency of the model. Conventional feature selection methodologies, traditional heuristic algorithms suffer in the result of poor scalability, huge search complexity, premature convergence of the model when working with the large feature spaces. In addressing these challenges, this research work proposes a Multi-Objective Hybrid Metaheuristic Optimization (MO-HMO) framework for optimizing the discriminative feature selection in the high-dimensional data. The proposed approach is the integration of adaptive exploration–exploitation strategies through cooperative population-based search mechanisms in effectively navigating the extensive complex high-dimensional feature space. In addition to this a multi-objective fitness function is also employed for simultaneously optimizing the conflicted criteria, including the maximizing of the feature relevance, minimizing the feature redundancy, and preserving the classification discriminative capability. Statistical relevance measures and stability-based criteria are incorporated into the optimization process to guide the selection of compact and informative feature subsets. Comprehensive experimental evaluations are conducted using standard benchmark functions and multiple high-dimensional datasets, including small-sample scenarios. The results demonstrate that the proposed MO-HMO framework consistently outperforms existing feature selection methods in terms of classification accuracy, dimensionality reduction, and convergence behavior. These findings confirm the effectiveness and robustness of the proposed approach for high-dimensional data analytics and real-world machine learning applications.

Ҳали таржима қилинмаган