Advancing Cervical Cancer Risk Prediction through Multi-Target Classification, SHAP Analysis, and Feature Reduction

This study presents a multi-target machine learning approach using Random Forest classifiers to predict cervical cancer risk factors based on the UCI Cervical Cancer Risk Factors dataset. The model concurrently predicts multiple clinical outcomes (Schiller, Cytology, Biopsy, and Hinselmann), enhancing predictive efficiency and addressing practical healthcare needs. SHAP (SHapley Additive exPlanations) analysis improves interpretability by quantifying the influence of individual features on each target variable. A feature importance assessment reveals that "Age" and "Number of Sexual Partners" significantly impact model performance, as evidenced by variations in AUC-ROC scores upon their exclusion. The model achieves high predictive accuracy, with AUC-ROC scores of 95.77% for Hinselmann target. This study integrates explainable AI techniques, feature relevance analysis, and visual analytics to support clinical decision-making and improve cervical cancer risk assessment.