Predictive Modeling of Metastatic Diagnosis Period in Breast Cancer using Machine Learning

The objective of this research is to utilize machine learning models to forecast the period of metastatic diagnosis in breast cancer patients, with a specific focus on triple-negative breast cancer (TNBC) data sourced from the WiDS Datathon 2024 dataset, which includes 19,000 records. This dataset encompasses a wide range of medical and socioeconomic variables that play a crucial role in comprehending healthcare disparities. Our assessment encompassed models such as Random Forest, XGBoost, and Support Vector Regression, with the Gradient Boosting Regressor (GBR) emerging as the most precise, achieving a Root Mean Square Error (RMSE) of 80.685. An analysis of feature importance revealed that the breast cancer diagnosis code, patient age, and geographic location (zip3) were the most significant predictors. Although feature importance was lower, payer type and socioeconomic factors also contributed to the model’s predictive capabilities. This study underscores the capacity of machine learning to capture both major and minor influences on metastatic diagnosis, thereby assisting in early detection and enhancing treatment strategies for TNBC patients.