Hybrid Graph Attention And Temporal Deep Learning Framework For Early Prediction Of Cervical Cancer Risk

Abstract

Cervical Cancer Remains One Of The Leading Causes Of Cancer-related Deaths Among Women Worldwide. Early Identification Of High-risk Individuals Is Essential For Timely Intervention And Improved Survival Rates. Traditional Machine Learning And Deep Learning Approaches Often Fail To Capture Complex Feature Relationships, Temporal Dependencies, And Provide Adequate Interpretability. This Study Proposes An Explainable Hybrid Graph Attention Network–Temporal Deep Learning (GAT-TDL) Framework For Cervical Cancer Risk Prediction Using Structured Clinical Data. The Framework Integrates Graph Attention Networks (GAT) For Modeling Feature Dependencies, Bidirectional Gated Recurrent Units (BiGRU) For Temporal Learning, And A Lightweight 1D Residual CNN For Feature Refinement. To Enhance Transparency And Clinician Trust, Explainable AI (XAI) Techniques Including SHAP And Integrated Gradients Are Incorporated. The Model Was Evaluated Using The UCI Cervical Cancer Dataset And Achieved An Accuracy Of 96.58%, Sensitivity Of 95.65%, Specificity Of 97.52%, And AUC-ROC Of 98.64%. Experimental Results Demonstrate That The Proposed Framework Outperforms Conventional Machine Learning Models While Providing Interpretable And Clinically Meaningful Predictions.

Keywords