A Machine Learning Based Approach for Prediction of Breast Cancer Patient Prognosis through Clinical Analysis

Abstract

Patient prognosis for cancer patients is a crucial aspect in the healthcare industry with researchers providing novel insights to doctors evaluating treatment options that have significant implications on patient lifestyle choices. By analyzing correlations of the genetic and clinical attributes for breast cancer patients, previous studies have utilized machine learning algorithms to predict the probability of patient survival based on a five-year timeframe. However, our project focuses on predicting a more specific label (overall survival months), utilizing the extensive dataset of an international breast cancer study, considering only the clinical attributes in scope. The application of the Multivariate Regression and Random Forest models was used to assess the relative importance of each clinical variable. The project results present the Random Forest model to be a better fit, accounting for 44% of the variance in the testing dataset. Further analysis with the expansion of other datasets would help improve the model accuracy.