GASE-XAI: A Genetic-Algorithm-Optimized Stacked Ensemble with SHAP and LIME Explainability for Breast Cancer Detection

(a) Problem Statement: Breast cancer is one of the most common cancers among women globally. Machine learning models have demonstrated high levels of accuracy in diagnosing breast cancer. However, these "black-box" models have not been clinically implemented because clinicians need transparent and interpretable models to make trustworthy decisions.

(b) Proposed Solution: The proposed study framework is coined GASE-XAI which stands for Genetic Algorithm-optimized Stacked Ensemble with Explainable Artificial Intelligence (AI). The Genetic Algorithm selects the most discriminative diagnostic features to reduce the dimensionality of the feature space. XGBoost, Random Forest, and Support Vector Machine classifiers are stacked and combined through a Logistic Regression meta-learner to yield predictions. SHAP and LIME techniques are leveraged for both global (population-level) and local (patient-specific) explanations.

(c) Significant Findings: The approach was tested on the Wisconsin Diagnostic Breast Cancer (WDBC) dataset with 569 patient examples, each with 30 diagnostic attributes. Feature selection through the Genetic Algorithm resulted in 63.3% fewer features (30 to 11). The predictive performance did not suffer, yielding cross-validated accuracy of 97.14% (95% CI: 95.49–98.80%) and 99.27% area-under-the ROC curve (AUC) on an independent test set. Statistical evaluation with Wilcoxon signed-rank tests, paired t-tests, and McNemar's tests demonstrated that there was no statistically significant difference in classification accuracy compared to six baseline classifiers (all p > 0.16). The proposed framework selects fewer input features while achieving comparable predictive performance.

(d) Applications: Computer-aided breast cancer diagnosis systems, clinical decision support systems, and early breast cancer screening programs are all potential applications for which GASE-XAI could be used. Two-level explainability can be used to increase clinicians' trust in the system, provide transparent medical decisions, aid in the adoption of AI models into real-world clinical workflows, and decrease computational burden by performing feature optimization.