Chi-Square Feature Selection and SMOTE-Balanced Ensemble Learning for Explainable Cervical Cancer Risk Prediction

• A compact cervical cancer risk-prediction framework
using training-data-only Chi-square feature selection to
retain 27 informative features.
• A Soft Voting ensemble of Logistic Regression, XGBoost, and CatBoost combined with SMOTE-based classimbalance handling.
• A multi-level SHAP analysis providing global feature importance, feature directionality, and individual prediction
explanations.
• An interactive research prototype that unifies modelbased risk prediction, probability and threshold information, and feature-level explanations within a single
interface.