Feature Selection Based Data Mining Approach for Coronary Artery Disease Diagnosis

Abstract

Cardiovascular diseases responsible for many deaths are very common and important health problems. According to World Health Organization, each year 17.7 million people die because of them. Coronary artery disease is the most important type of cardiovascular diseases that cause serious heart problems in patients, affecting the heart’s function negatively. Being aware of the important attributes for this disease will help field-specialist in the analysis of routine laboratory test results of a patient coming internal medicine or another medicine unit except for the cardiology unit. In this study, it is aimed to determine the significance of attributes for coronary artery disease by utilizing Stability Selection method. In experiments, the attributes; ‘Age’, ‘Atypical’, ‘Blood pressure’, ‘Current smoker’, ‘Diastolic murmur’, ‘Dyslipidemia’, ‘Diabetes mellitus’, ‘Ejection fraction’, ‘Erythrocyte sedimentation rate’, ‘Family history’, ‘Hypertension’, ‘Potassium’, ‘Nonanginal’, ‘Pulse rate’, ‘Q wave’, ‘Regional wall motion abnormality’, ‘Sex’, ‘St Depression’, ‘Triglyceride’, ‘Tinversion’, ‘Typical chest pain’ and ‘Valvular heart disease’ were found important for each sub-dataset. Besides, the performances of four traditional machine learning algorithms were evaluated to detection of this disease. Logistic Regression algorithm outperformed others with %90.88 value of accuracy, 95.18% value of sensitivity, and 81.34% value of specificity.