Predicting Myocardial Infarction using Data Mining and a Two Stage Feature Selection Method

Myocardial infarction is among the most usual all over the world. Applying data mining methods for making a model to predict occurrence of myocardial infarction is very helpful. According to medical references, the most reliable and common method for detecting the blockage in heart arteries is Angiography, which is costly and has side-effects. Thus, using a method for early detection of the myocardial infarction, and therefore, eliminating the need for Angiography is very important. The purpose of this study is to present a model for early predicting myocardial infarction using classification methods.In this study a dataset consist of 519 patient with 52 feature is used. The data is obtained from the information of the visitors to Shahid Madani Specialized Hospital of Khorram Abad, Iran. The features of the data set are divided to 3 groups: Demographic features, Symptoms and physical examination, Laboratory features and ECG related and Arteries. In this study, the algorithms Naïve Bayes, Support Vector Machine, Artificial Neural Network and K-Nearest Neighbors are used for predict myocardial infarction. Besides, a two stage feature selection method, include weight by SVM and Genetic Algorithm, is applied to the data set. The Highest accuracy of the classification algorithms, after feature selection, belong to K-NN which provide 97.69%. The result showed that using new data such as Troponin I and CRP, and also applying two stage feature selection method lead to better performance.

Keywords:

Myocardial Infarction , Data mining , Classification , Genetic Algorithm , K-Nearest neighbors , Multi-layer perceptron , Naïve Bayes , Support Vector Machine