Comparative analysis of the accuracy of an automated machine learning model for detecting cardiovascular diseases

Abstract

cardiovascular diseases (CVD) are widespread among patients with chronic non-communicable diseases and are one of the leading causes of mortality in the population, including those of working age. The development of patient-oriented systems for early detection of cardiovascular diseases using machine learning models is a promising direction that integrates medical knowledge and information intelligent technologies for medical decision support systems. To simplify and speed up the process of developing a specific solution based on a wide variety of machine learning models, the field of automatic machine learning (AutoML) is actively developing. The article provides a comparative analysis of the accuracy of an AutoML model created using the AutoGluon-Tabular library. The accuracy comparison was carried out in two directions: in relation to three scenarios for preprocessing data from patients with CVD and in relation to the basic machine learning models contained in the AutoGluon-Tabular library. A comparative analysis on the open UCI database showed that the accuracy of the AutoML model in identifying cardiovascular diseases varies from 87.41% to 95.65%, with the maximum accuracy obtained in the scenario with Z-normalization of the original data, and the minimum accuracy - when using data preprocessing algorithm built into AutoML.