Optimizing Heart Disease Prediction: A Comparative Study of Machine Learning Techniques

Abstract

Healthcare is an important sector which needs continued modernization with technology-based state-of-the-art system. Main focus of this paper is applying artificial intelligence, especially machine learning techniques to build heart disease prediction system. In this sector, diagnosis of diseases like heart disease by thorough test reports investigation of doctors is crucial, challenging and time-consuming. In addition to doctor’s investigation, artificial intelligence techniques can assist and alleviate healthcare hassles and challenges. This paper presents prediction system for heart disease using nine machine learning methods. The experimented heart disease dataset has 918 sample and 11 features (key features include Age, and Fasting Blood Sugar, correlation between Maximum Heart Rate, Gender, Chest Pain Type, Exercise Angina, peak slope of exercise ST segment).  Among the methods experimented on heart disease data, Categorical Boosting (CatBoost) and Light Gradient Boosting Machine (LGBM) were outperformers among the methods with highest accuracy scores and Recall scores. Higher Recall scores contribute to correctly predict the true positive values or correctly predicting true heart disease patients which is very important in sectors like healthcare. Also, CatBoost and Random Forest were best performing methods in 10-fold cross-validation test. All these experimented results reflect that artificial intelligence or specifically machine learning algorithm-based prediction system for heart disease can be a very assisting tool for physicians and overall healthcare system. This research encourages further investigation with more and large datasets. Continued research and availability of more large datasets can improve prediction accuracy to higher satisfaction level of doctors which can help develop predictive ecosystem.