Ensemble Learning for the Survivability Prediction of Breast Cancer Patients Using METABRIC and SEER Datasets

Abstract

Breast cancer is the most life-threatening disease among women worldwide. Nowadays, clinical and genomic data in breast cancer datasets are analyzed to identify the incidence and impact of the disease. This paper proposes an analysis on the clinical breast cancer datasets, METABRIC and SEER for the prediction of survivability of breast cancer patients through ensemble classifiers. It attempts to identify informative independent features for those classifiers using filter methods, Mutual Information and Chi-Square and wrapper methods, Sequential Forward Selection, Sequential Backward Selection and Recursive Feature Elimination in the feature selection sequence. An analysis on the performance of ensemble techniques namely, bagging, boosting and stacking shows that the ensemble techniques produce higher accuracy when compared to the accuracies produced by the base classifiers. Also, the stacking ensemble model produces higher accuracies of 99.2% and 99.8% than other ensemble techniques on METABRIC and SEER datasets respectively. The SBS wrapper method aids higher accuracy to all classifiers than SFS and RFE.