Predicting & Diagnosing Heart Disease Patients Using Multinomial Logistic Regression

Abstract

Heart disease is one of the most common diseases and the principal cause of sudden death nowadays. According to the Iraqi Ministry of Health, 28.83% of deaths in Iraq are caused by some kind of heart disease. This is motivated by the worldwide increasing fatality of heart disease (HD) patients each year and the availability of statistical methods that can be used to extract convenient knowledge from data collected from patients, to aid physicians in diagnosing heart disease. A statistical model was developed using multinomial logistic regression (MLR) to predict heart diseases using a training dataset of 201 patients collected from Alkarama Teaching Hospital. The data were statistically analyzed to reduce bias and assess how much the attributes explain the diversity of the dependent variables, which resulted in 95.8%. The overall prediction model had an accuracy of 94.5%. The researchers concluded that the MLR model has high accuracy in predicting heart disease, and that echocardiographic findings of hypokinesia, chest pain, and electrocardiographic results are the top three attributes, in terms of accuracy, for diagnosing heart disease.