The Utilization and Comparison of Artificial Intelligence Methods in the Diagnosis of Cardiac Disease

Abstract

Today a significant amount of human mortality is because of cardiac disease. These mortality could be reduced considerably by diagnosis on early stages. In this study we propose an artificial intelligence based early diagnosis system for cardiac disease prediction. For the research we utilized Cleveland and Z-Alizadehsani datasets. For Cleveland database which contains 76 attributes, 13 attributes selected in order to predict heart disease presence. For Z-Alizadehsani database which contains 55 attributes, all attributes are utilized for prediction. System implements not only basic classifiers as Naïve-Bayes, Linear Regression, Polynomial Regression, Support Vector Machine (SVM) but also ensemble classifer Random Forest and complex models like artificial neural network based multilayer perceptron. On cardiac disease prediction two cross validation techniques employed. Firstly 20 experiments processed for each method by utilizing holdout cross validation technique. Secondly K-fold (10 fold) cross validation is applied for all methods. Multiple Linear Regression with holdout cross validation has achieved best results as 0.91 accuracy for Cleveland dataset and 0.91 for Z-Alizadehsani dataset. For these two datasets when K fold is utilized 0.93 accuracy score achieved for both. Best result is obtained as 0.97 accuracy by SVM method with Z-Alizadehsani dataset. Generally it is observed that K fold method has better results than Holdout method. Detailed and comparable results of experiments are given in tables. Illnesses could be detected correctly in early phases by integrating these models to health systems like hospital otomations. The proposed system could be presented as continous learning web service to health automation systems.