Comparison of Traditional Cox Proportional Hazard Model with Machine Learning Models for Survival Analysis in Predicting Risk of Death due to Heart Failure

Risk of death due to heart failure can depend on different biological or anatomical details of the patients. In this paper we used a dataset containing medical records of 299 patients who were monitored over a certain period. Though Cox Proportional Hazard (CPH) Model is the most conventional approach while analyzing survival data, machine learning (ML) models are also being used recently. The problem with these ML methods is that they do not take ‘censoring’ into account. To incorporate censoring, especially right censoring, here in this article we have used Random Survival Forest Model, Gradient Boosted Model and Survival Support Vector Machine to predict the risk of death due to heart failure and compared their performances with traditional CPH model by Harrell’s Concordance index and time dependent AUC. At the end of the study, it is seen that traditional CPH model performs better than rest of the ML techniques.