Feature Selection Optimization in High-Dimensional Data: A Hybrid Approach Based on Botox Optimization Algorithm and Q-Learning

Feature selection is one of the fundamental and practical problems in machine learning, playing a critical role in improving model performance. Datasets used for training learning algorithms often contain numerous features, many of which are redundant and can negatively impact model accuracy. Therefore, identifying and selecting the most relevant and influential features to enhance model performance is a key step. The goal of feature selection is to reduce the dimensionality of features while preserving the most effective ones, ensuring that the model achieves maximum accuracy with the minimum number of features. This process can be formulated as an optimization problem and solved using optimization algorithms. This paper proposes a novel hybrid approach based on the Botox Optimization Algorithm (BOA) and Q-learning. The BOA efficiently balances exploration and exploitation, making it well-suited for optimization tasks. However, it may get trapped in local optima and fail to sufficiently refine the internal details of solutions. To address this limitation, after the initial convergence of BOA to promising regions, the solutions are further refined using the Q -learning algorithm. By defining states and actions, Q-learning improves the details of solutions and enhances final accuracy. The performance of the proposed method is evaluated on datasets from the UCI repository and compared with genetic algorithms, particle swarm optimization, and the standard BOA. The obtained results indicate the superiority of the proposed method in terms of accuracy and efficiency in feature selection.