Abstract

This paper studies the application of survival machine learning models in management for outcome prediction based on the medical literature. Twenty survival models and over ten survival machine learning algorithms were analyzed to find their key advantages and disadvantages. In the first half of this study, we examine and evaluate the most prevalent models in terms of their similarities and differences, as well as their data types and evaluation strategies. We also highlight the concepts that all machine learning algorithms for survival analysis must adhere to. Four machine learning algorithms from each family (trees, multitask, kernel, and deep network) were used to analyze a breast cancer dataset and two additional simulated datasets using the R coxed package. The results indicate how machine learning algorithms might be used to recommend medicines and improve population health by analyzing survival. Moreover, we establish the ideal approaches to use based on more than twelve limitations, such as suppressed data.

Keywords: Survival analysis, Statistical inference, Survival machine learning, Time to event analysis