Detecting Cardiac Ailments using Machine Learning

According to World Health Organization (WHO), cardiovascular diseases are world's leading cause of deaths with approximately 32% (17.9 million) of all deaths every year. Early detection of heart related illness may prove beneficial in decreasing the deaths worldwide. Currently, detection is done by a doctor based on the test reports and lab results. Patients tend to consult with multiple doctors just to decrease the human error coefficient which costs them a lot of money and takes time. To tackle this problem, machine learning based techniques are getting popular to provide non-invasive solutions for detection of heart diseases. In this paper, we propose a machine learning based disease detector which can be used to check whether a patient has some kind of heart disease or not. The proposed model uses a Gradient boosting classifier with a Grid Search algorithm to select the best hyper-parameters. The model utilizes information from patient's test reports such as chest pain type, cholesterol, condition of thalassemia etc., and classifies the patient as a probable heart disease patient or non-probable. The data undergoes multiple feature transformations based on H20.ai auto feature transformations and multiple feature selection algorithms to select the best features. The model has achieved an accuracy of 93%, recall of 94.2% and precision of 92.5% on UCI Heart Disease Dataset.