Performance Evaluation of Machine Learning & Robust Estimation for Parameters in Survival Analysis

The goal of survival analysis as a branch of statistics is to analyze data, model it and estimate the time it takes for a certain event to occur. The event outcomes for the existence instances are not observable after a specific amount of time or instances that do not experience any events in the period of observation present one of the main challenges in this case. The best way to deal with this-so called censoring is to use survival analysis techniques. Furthermore, numerous algorithms of machine learning have been modified to handle this type of censored data as well as other difficult issues that come up in real world data. In this paper we provide a comprehensive comparison between parametric survival models by using Buckley James estimation method, Robust parametric survival models by using Buckley James estimation method with Tukey’s Biweight function and machine learning method to build a survival tree for (Exponential, Weibull, Weibull Three Parameter and Log-normal) parametric models. Based on these comparisons the best models were detected by depending on the smallest AIC and BIC.