Hybrid Machine Learning-Based Survival Analysis Model for Predicting Breast Cancer Recurrence

Hybrid Machine Learning-Based Survival Analysis Model for Predicting Breast Cancer Recurrence

Monireh Shahmoradi1

1) Master of Mathematical Statistics, Yasuj University, Yasuj, Iran.

IOI Code : AAGF-AAHGZC
Publication : 9th International Conference on Applied Researches in Science & Engineering (carseconf.com/9th)
Abstract :
    Breast cancer recurrence represents a major clinical challenge in oncology, particularly within Iran, where increasing incidence rates and heterogeneous treatment outcomes demand more accurate survival prediction tools. Traditional statistical models such as the Cox proportional hazards approach, while foundational for recurrence analysis, often fail to capture nonlinear interactions inherent in multidimensional clinical datasets. In response, the present study develops and validates a hybrid machine learning–based survival analysis model designed to improve recurrence forecasting accuracy and interpretability among Iranian breast cancer patients. The study employed a retrospective quantitative design using anonymized clinical records from major oncology centers across Tehran, Shiraz, and Mashhad, encompassing data collected between 2021 and 2025. The hybrid framework integrated Cox regression for baseline hazard estimation with ensemble algorithms—Random Forest Survival and XGBoost‑Survival—to model nonlinear feature interactions. Evaluation metrics included the concordance index (C‑index), time‑dependent AUC, and calibration accuracy, computed through Python and R analytical environments. Results revealed that the hybrid model achieved a C‑index of 0.87, outperforming both the standalone Cox model (0.74) and single machine learning models (0.84). It maintained high calibration fidelity across follow‑up intervals and effectively stratified patients into high, moderate, and low recurrence risk groups. Key predictive factors included lymph node involvement, tumor grade, and hormone‑receptor status, confirming biological robustness while offering interpretable insights through SHAP analysis. The findings substantiate that combining statistical inference with machine learning adaptation yields a balanced, transparent, and high‑performance survival estimation tool suitable for Iranian clinical contexts. The study concludes that hybrid survival modeling provides a reliable methodological advance capable of supporting precision oncology by integrating epidemiological structure, computational intelligence, and clinical applicability. Overall, the framework not only enhances predictive capability but also contributes to developing a nationally adaptive analytical ecosystem for breast cancer prognosis in Iran, paving the way for future prospective validations and real‑time implementation in hospital decision‑support systems.

Keywords : breast cancer recurrence survival analysis hybrid machine learning cox model iran