An Explainable Hybrid GWO-LightGBM Model for Breast Cancer Diagnosis Using SHAP Interpretation

Abstract

The rapid growth of machine learning and data mining has has transformed the field of medical diagnostics. Automated systems have been developed that can identify complex disease patterns with high accuracy. However, most current models still include redundant and correlated features, which increase computational cost, reduce interpretability, and may lead to overfitting. This research proposes a hybrid diagnostic system that combines Grey Wolf Optimization (GWO) for feature selection with the Light Gradient Boosting Machine (LightGBM) classifier to enhance both accuracy and interpretability. The proposed GWO–LightGBM model was evaluated using the Breast Cancer Wisconsin dataset. The framework successfully reduced the number of input features from 30 to 12. In addition, it achieved a test accuracy of 97.37% and a cross-validation accuracy of 98.02% ± 0.02, outperforming the baseline LightGBM model trained on all features. Furthermore, the proposed model reduced training time by 25% and demonstrated statistically significant improvement (p < 0.05). Moreover, SHAP analysis revealed that the selected features were biologically meaningful, enhancing the model’s transparency and trustworthiness. The proposed approach demonstrates that integrating feature selection with LightGBM and explainable artificial intelligence techniques can produce fast, interpretable, and reliable diagnostic models for healthcare applications