Explainable Ensemble Learning for Classification Breast Tumor Using the Wisconsin Diagnostic

Breast tumor classification is a binary classification problem that can benefit from machine learning, but high predictive performance alone does not explain why a model produces a particular decision. This study develops an explainable ensemble-learning workflow using the public Wisconsin Breast Cancer Diagnostic (WDBC) dataset. The dataset contains 569 samples, 30 numerical features, and two diagnostic classes. Three tree-based ensemble models, Random Forest, Extra Trees, and XGBoost, were trained using an 80:20 stratified split and evaluated over five random seeds using accuracy, precision, recall, F1-score, and ROC-AUC. SHAP was subsequently applied to the strongest model to identify globally influential features. Across five repetitions, Extra Trees and XGBoost obtained the highest mean accuracy of 95.96%, while Extra Trees achieved a mean F1-score of 96.85% and ROC-AUC of 99.43%. On the fixed seed-42 test set, Extra Trees achieved 95.61% accuracy, 95.60% weighted F1-score, and 99.47% ROC-AUC. SHAP analysis indicated that worst perimeter, worst radius, worst concave points, worst area, and mean concave points were among the most influential features. The results show that combining ensemble classification with post-hoc explanation can provide both predictive evaluation and interpretable feature-level evidence, while the findings should be treated as benchmark results rather than a clinical diagnostic system.