Prediction of cholesterol level in patients with myocardial infarction based on medical data mining methods

Abstract

Myocardial infarction (MI) is a significant reason for death and disability over the world and might be the first sign of coronary artery disease. The current study was carried out to predict the cholesterol level in patients with MI using data mining methods, artificial neural networks (ANNs) and support vector machine (SVM) models. The data of 596 patients, who had been diagnosed with segment elevation MI were analysed in the present study. The retrospective dataset including gender, age, weight, height, pulse, glucose, creatinine, triglyceride, high-density lipoprotein, and low-density lipoprotein was used for predicting the cholesterol level. Correlation based feature selection was applied. Multilayer perceptron (MLP) ANNs and SVM with radial basis function kernel were used for the prediction based on the selected predictors. The performance of the ANNs and SVM models was evaluated on the basis of correlation coefficient and mean absolute error. The estimated correlation coefficients observed and predicted values were 0.94 for ANNs and 0.88 for SVM in training dataset (n=376), and 0.95 for ANNs and 0.90 for SVM in testing dataset (n=160), respectively. ANNs and SVM models yielded mean absolute error of 7.37 and 14.18 in training dataset, and 7.87 and 14.71 in testing dataset, consecutively. The results of the performance evaluation showed that MLP ANNs performed better for the prediction of cholesterol level in patients with MI in comparison to SVM. The proposed MLP ANNs model might be employed for predicting the level of cholesterol for MI patients in clinical decision support process.