Optimizing Neural Network Performance to Predict Coronary Heart Disease

Coronary heart disease is a disease that causes the death of most people in the worldwide. According to the WHO, 9.43 million people died from ischemic stroke in 2018.Residents of some CIS countries (Belarus, Kazakhstan, Kyrgyzstan, Russia and Ukraine) have a higher cardiovascular mortality rate at the age of 55-59 than the French at the age of 75-79. Premature mortality rates in women in European countries are lower than in men, but have similar dynamics [12], [13].The constant increase in the amount of information in cardiology makes the development of new methods for data analysis urgent. Using existing risk assessment approaches, it is impossible to predict about half of the episodes of acute coronary syndrome. Big data machine learning can lead to better diagnostic and treatment outcomes at a lower cost. The inductive approach allows you to identify patterns arising from data analysis and develop algorithms that can learn on their own. Although machine learning models for cardiovascular risk assessment are superior to traditional methods, to date, no large-scale machine learning studies have been conducted to prove a predictive role in the general population using routine clinical data. In addition, there is no clear recommendation which of the algorithms will work better in a given situation. The use of an empirical approach when choosing a machine learning method and the principle of "black box" make it difficult to conduct large-scale studies and implement machine learning methods in clinical practice. This work is devoted to the study of the optimization of neural networks using a correlation analysis of signs to predict the possibility of coronary heart disease. As a result of the analysis of literary sources, the shortcomings of previously developed software tools were identified, seriously limiting the use of these programs in the medical field.