A Comparative Study for Survival Prediction of NKI Data Using Statistical and Machine-Learning Approaches

Abstract

The main aim of the study is to compare both statistical and machine learning algorithms in survival analysis prediction, taking into account the dimensionality of the dataset. Performance criteria like the concordance-censored index have been employed in the context of 5-fold and 10-fold cross-validation. The dataset utilized was the Netherlands Cancer Institute (NKI). Two machine-learning algorithms (eXtreme Gradient Boosting and Fast Survival Support Vector Machine) and one statistical technique (Cox proportional hazard model) were employed. The results showed through the application that, as features utilized increase, XGBoost surpasses other algorithms. However, as the number of selected features decreases drastically, the statistical analysis and the Fast Survival Support Vector Machine are both better at survival prediction.