Classification Based on Event in Survival Machine Learning Analysis of Cardiovascular Disease Cohort

Abstract

The goal of this paper is to evaluate supervised learning classification models in a survival analysis problem of cardiovascular patients with a significant cured fraction. A sample include 919 patients referred to Sulaymaniyah Cardiac Hospital (including 365 females and 554 male) were followed up for a maximum of 650 days in 2021 to 2023. In the sample, 162 patients (17.6%) during research time died. Cure fraction in this cohort was confirmed based on the Mahler and Zhu test (P<0.01), so for this case several machine learning classifications are applied to find the best patient status prediction procedure. Using several machine learning algorithms, the patients were classified into alive and dead. The results based on several indicators were almost the same. But in general, the random forest was identified as the best method in most of the indicators and of course with the Area under ROC = 0.934. The just weakness point of this method is that it did not work relatively well in the correct diagnosis of the deceased, and instead, the SVM with FP Rate = 0.263 performed better. After that, logistic and simple regression with Area under ROC = 0.911 and 0.909 showed better performance than others.