Comparing the Performance of Traditional Machine Learning and Deep Learning Algorithms for Breast Cancer Survival Prediction
Navaneeth Ranjit
Affiliation: Private International English School
IJSCAR Vol. 3, Issue 1 (2026) · pp. 13–23
Abstract
Breast cancer remains the most commonly diagnosed cancer among women worldwide with disparities in outcomes influenced by geographic and socioeconomic factors. In this study we address the challenge of predicting breast cancer survivability specifically in scenarios with limited data. Using the METABRIC dataset which comprises clinical and genetic attributes from 1980 primary breast cancer samples we performed binary classification to predict patient survival outcomes. We explored a range of machine learning and deep learning models including 13 traditional ML algorithms and 6 deep learning architectures. In addition we implemented a time-to-event analysis using a Cox proportional hazards pipeline with Elastic Net regularization and deep survival models such as DeepSurv and DeepHit. A soft voting classifier combining LightGBM CatBoost and AdaBoost trained and tested on clinical attributes provided the best performance while deep learning models underperformed likely due to the limited dataset size. Overall we observed that the models performed better when trained on clinical features rather than using genetic attributes suggesting that clinical indicators remain more predictive of breast cancer survivability.
Keywords: Breast Cancer Survival, Machine Learning, Deep Learning, Ensemble Models, Tabular Data, METABRIC, Genetic Attributes, Clinical Attributes