Abstract

This study introduces a hybrid technique that combines the Crow Search Algorithm (CSA) with Proximal Policy Optimization (PPO) from reinforcement learning for feature selection in large datasets. The integration of PPO allows the algorithm to adapt and learn during its execution, thereby improving efficiency and accuracy in identifying relevant attributes. We evaluate the effectiveness of this hybrid approach using the Parkinson’s Disease Classification (PDC) dataset. Our computational results demonstrate promising improvements in solution quality, convergence speed, and reduction in execution time, especially in high-dimensional environments. This work not only highlights the feasibility of combining bioinspired algorithms with reinforcement learning techniques but also opens new avenues for optimizing feature selection processes in machine learning.