Comparative Performance Analysis of Random Forests against AutoPrognosis for predicting Coronary Heart Disease Risk and Metabolic Syndrome: A Retrospective Cohort Study

¹ School of Computer Science and Applied Mathematics, University of the Witwatersrand, 1 Jan Smuts Ave, Braamfontein, Johannesburg, South Africa
² School of Computer Science and Applied Mathematics, University of the Witwatersrand, 1 Jan Smuts Ave, Braamfontein, Johannesburg, South Africa

^* Corresponding author: paulinangcayiya@gmail.com

Abstract

Cardiovascular Disease (CVD) is the leading cause of mortality worldwide. Amongst them, Coronary Heart Disease (CHD) is the most common type of CVD. The consequences of the presence of CVD risk factors often manifest as Metabolic Syndrome (MetS). In this study, data from the Framingham Heart Study (FHS), consisting of 4240 records and 17 variables, was used to build two types of 10-year CHD risk prediction models based on Random Forests (RF) and AutoPrognosis. The Framingham Risk Score model (AUC-ROC: 0.633) was used as a baseline model for performance evaluation. Results showed that the RF model with optimized hyperparameters had the best performance (AUC-ROC: 0.728). Furthermore, a dataset of 7821 records and 77 variables from the National Health and Nutrition Examination Survey (NHANES) was used to assess the predictive performance of RF against AutoPrognosis for determining the presence of MetS. The RF model with optimized hyperparameters had the best performance (AUC-ROC: 0.851). The performance of RF against AutoPrognosis on different sample sizes of data, ranging from 100 to 4900, was tested. The RF model with optimized hyperparameters had the best overall performance, followed by AutoPrognosis with an ensemble pipeline, then AutoPrognosis with a single pipeline and finally the RF model with default hyperparameter values.