Hybrid GraphSAGE–ANN for Lung Cancer Risk Prediction Using Patient Similarity Graphs

This paper proposes a novel Hybrid GraphSAGE–ANN model that fuses graph-based patient-similarity representations with non-relational feature learning for lung cancer risk prediction. Benchmarked against ten baseline and hybrid GNN architectures on a 5,000-patient dataset under a unified, leakage-safe evaluation protocol, the proposed model achieves the best overall performance (98.80% accuracy, 97.53% F1-score, 99.96% ROC-AUC) and the strongest clustering-agreement scores among all graph-based models, while uniquely offering an interpretable patient-similarity graph for clinical cohort analysis.