Hybrid Evolutionary Optimization Techniques for Feature Selection and Classification Applications

Abstract

Two of the most common tasks in machine learning, feature selection and classification have a strong bearing on the model performance, computational efficiency, and interpretability. Nevertheless, the high-dimensional data is usually filled with redundant, irrelevant, and noisy features that can deteriorate classification and raise the level of computational complexity. This study proposes a hybrid evolutionary optimization algorithm to feature selection and classification tasks, using the advantages of various methods of evolutionary computation to find optimum feature sets and improve predictive performance. The suggested strategy is a combination of the evolutionary algorithm global exploration capabilities and adaptive optimization strategies to attain an effective exploration and exploitation balance. Advanced classification models then use the selected features to make better predictions, minimize training time, and improve the ability to generalize. The framework is tested on benchmark data on various application areas and judged by various performance measures, including accuracy of classification, rate of feature reduction, computational cost, and robustness of the model. The experimental findings indicate that the hybrid evolutionary optimization approach is effective in removing irrelevant features, and the important discriminative information, which results in a better classification performance as compared to the traditional feature selection algorithms and independent optimization algorithms.