XAI-CancerNet: An Explainable Artificial Intelligence Framework Using XGBoost and SHAP for Breast Cancer Diagnosis and Clinical Decision Support

Abstract

Breast cancer is one of the most common and deadly cancers worldwide, with more than 2.3 million new cases reported each year. The precise and timely categorization of breast cancers as malignant or benign is essential for improving patient survival rates and minimizing unnecessary invasive procedures. This research presents XAI-CancerNet, an innovative explainable artificial intelligence (XAI) framework that integrates Extreme Gradient Boosting (XGBoost) with SHAP (SHapley Additive exPlanations) for post-hoc evaluation. This combination yields precise, lucid, and intelligible breast cancer predictions for healthcare practitioners. Utilizing the UCI Breast Cancer Wisconsin (Diagnostic) dataset, which contains 569 samples and 30 morphological parameters of cell nuclei, the proposed methodology attains a test accuracy of 98.6%, a precision of 98.1%, a recall of 99.0%, an F1-score of 98.2%, and a ROC-AUC of 99.4%. This performance exceeds that of all other assessed advanced methodologies from 2021 to 2026. The SHAP analysis determines the five most critical features for diagnosis as the worst concave points, worst perimeter, worst radius, mean concave points, and worst area. This resource offers oncologists outstanding graphic support consistent with established competence in the domain. A thorough 5-fold stratified cross-validation indicates that the model operates efficiently among patients of diverse ages and demographics. The XAI-CancerNet system associates anticipated accuracy with clinical reliability. It functions as an instrument for computer-aided diagnostic (CAD) systems and decision-support platforms in oncology.