Feature Selection Algorithm Based on Q-learning and Artificial Lemming Algorithm for Medical Data

In medical diagnosis, accurate disease identification is paramount. Enhancing model accuracy significantly improves diagnostic outcomes. Medical data is redundant and the relationships among features are complex. Feature selection algorithms can remove redundant features and noise to improve the performance of the model. This paper proposes QALA, a novel hybrid feature selection method for medical data that integrates ReliefF-based Q-learning and artificial lemming algorithm (ALA). The approach first employs ReliefF with Q-learning-optimized thresholds for preliminary feature screening, then applies an improved ALA incorporating ReliefF-guided population initialization. A hybrid search strategy combining historical optimal positions with sine cosine optimization (SCA) to enhance solution space exploration. Experimental results on multiple biomedical datasets demonstrate that QALA outperforms 10 state-of-the-art metaheuristic algorithms in feature selection performance.