Predicting Severity from Electronic Health Records of Leprosy Patients using Ensemble Learning

Abstract

Electronic Health Records (EHRs) are speedily being enforced by healthcare providers in recent years. Leprosy is a specially listed neglected tropical disease that continues as a major health problem in India. The delay in the diagnosis can lead to increase disability rate among patients. This paper intends to identify various risk factors from EHRs by applying ensemble machine learning techniques. The EHRs are included with the first sign of symptoms and various diagnosis details of leprosy cases. This information is used to determine the severity of leprosy cases and classify them into 3 categories, namely mild, moderate, and severe. To predict the severity, AdaBoost and XGBoost ensemble classifiers are applied in this paper. The performance of these classifiers is compared with Classification and Regression Trees (CART) and Random Forest (RF) techniques. The results show that AdaBoost gives with 97% accuracy and 97% precision. XGBoost gives 97% accuracy and 99% recall.