MEDPREDICT: Smart Predictive Healthcare Model for Early Detection of Heart and Diabetes Diseases Using Machine Learning

Abstract

Heart disease and diabetes are the leading causes of mortality across the world today. According to the WHO, heart disease is the leading cause of death; diabetes is the silent doubling risk factor of heart failure and stroke. Since these two diseases are inextricably connected, physicians must have only one test which examines the two, yet most procedures are performed independently. In this paper, I present a smart model called MedPredict, which would be able to predict the risk of heart disease and diabetes simultaneously using combined data. MedPredict combines machine-learning ensembles with sophisticated optimization techniques. The initial step involves first identifying significant data features with 2 algorithms: Genetic Algorithm to explore a wide range of potential solutions, and Particle Swarm Optimizer to refine the most promising solutions, which means preventing overly complicated models. To classify patients, it uses the predictions of three models (XGBoost, Random Forest and Support Vector machine) and averages the probability of the models. This stabilizes the predictions and they are accurate on new patients. The system is also designed to be an IoT -cloud system, and thus will be capable of receiving real-time health data provided by wearable devices and sending it to a cloud engine to be analyzed. The combined heart and diabetes (UCI Heart & Pima) tests indicate that MedPredict is also more accurate (95.2) than the individual models and can be scaled up to personalized health care.