Advancing Cervical Cancer Risk Stratification via Ensemble Learning Models Integrated with SHAP-Based Interpretability Methods

Cervical cancer remains a leading cause of cancer-related mortality among women globally, particularly in regions with limited access to regular screening. Early detection improves prognosis, yet traditional methods such as Pap smears and HPV testing are often unavailable in low-resource settings. In this work, we present a novel machine learning framework for cervical cancer risk prediction, leveraging a Stacking ensemble of Gradient Boosting (GB), Random Forest (RF), XGBoost (XGB), and Support Vector Machines (SVM). To address class imbalance in the dataset, we apply the Synthetic Minority Over-sampling Technique (SMOTE), while Recursive Feature Elimination (RFE) identifies the most relevant clinical and demographic features. Model interpretability is ensured through SHapley Additive exPlanations (SHAP), providing insights into feature contributions. Our approach achieves a state-of-the-art accuracy of $\mathbf{9 9 . 0 0 \%}$, outperforming individual models and demonstrating the effectiveness of ensemble learning for early diagnosis. This work presents a scalable, interpretable, and clinically viable solution for improving cervical cancer screening in resource-constrained environments, setting a new benchmark in predictive modeling for medical diagnostics.