Ensemble Learning Approach Reveals Significant Clinical Attributes from Real-World Breast Cancer Cases

Abstract

Breast cancer has become on of the leading causes of death in Indonesia. This study contributes to global efforts to combat breast cancer by improving patient outcome prediction accuracy. This study employed ensemble learning techniques such as Random Forest, XGBoost, and LightGBM. The results of the study demonstrates LightGBM's superior performance (accuracy=85%, ROC-AUC=81%, AUPR=85%). Notably, all three algorithms identify key clinical attributes: "Relapse Free Status (Months)", "Overall Survival (Months)", "Nottingham Prognostic Index", and "Lymph Nodes Examined Positive". LightGBM uniquely highlights "pam50_LumA" as significant, suggesting reduced fatality risk for Luminal A subtype patients, while others prioritize "Tumor Size". This research lays groundwork for intelligent systems to predict breast cancer outcomes, potentially transforming patient care and clinical practice.

Keywords

• Breast cancer
• Ensemble learning
• Machine learning
• Patient's outcome prediction
• Clinical data analysis