Kalp hastalık risk tahmini için Python aracılığıyla sınıflandırıcı algoritmalarının performans değerlendirmesi

Abstract

Heart diseases are one of the biggest health problems of today. Early diagnosis for the disease can prevent early deaths. For this purpose, by using 13 independent variables in the data set obtained from the Kaggle database, people with low probability of heart disease and people with excess were tried to be distinguished. Seven classification algorithms were used in the study, namely support vector machines (SVM), k-NN, decision trees, linear discriminant analysis (LDA), Gausian Naive Bayes (GNB), Gradient Boosting (GB) and Random Forest (RF). Random forest was the algorithm that made the best estimation of the study according to the values of specificity (100%), Matthews correlation coefficient (0.90), Fowlkes-Mallows index (0.82), F1 score (89.7%) and accuracy (90.2%). There was no statistically significant difference between the groups in fasting blood glucose and it was found to be the least important among the features. No significant performance change was observed in the classification processes made by removing this feature. Only the processing times are slightly shorter. This study will help predict heart disease as it will support early diagnosis.