Class imbalance handling and metaheuristic optimization of XGBoost for breast cancer detection

The main aim of this experiment was to optimize breast cancer detection on the Wisconsin Diagnostic Breast Cancer (WDBC) dataset with the help of (PSO) to hyper-optimize the XGBoost classifier. The default XGBoost model was found to have an accuracy of 96.49% percent with high levels of precision and recall between the benign and the malignant classes. The model performance increased and it reached an accuracy of 97.37%, precision of 0.98%, and recall of 0.96% with an overall F1-score of 0.97% after the PSO optimization, which is the confirmation of the effectiveness of the optimization process. SHAP was used to analyses global feature importance to ensure interpretability, to identify the most important factors affecting predictions, and LIME to explain one case at a time, providing clinical-level transparency. A comparative analysis against most recent 2025 publications also indicated that the PSO-tuned XGBoost model outperforms most state-of-the-art models in accuracy and F1-score.