Explainable Machine Learning Techniques For Predictive Analysis Of Chronic Diseases Using Electronic Health Records

Abstract

Diabetes, cardiovascular diseases and chronic kidney diseases are among the chronic diseases responsible for a significant burden of disease and increasing mortality, complications and healthcare costs at the global level. Electronic Health Records (EHRs) have been growing very rapidly and have been used to build a data-driven predictive healthcare system to assist the early diagnosis and personal treatment plans. Many machine learning (ML)-based predictive models, however, are black box models, obscure – and not trusted – by clinicians when it comes to automated decision making. In this study, an explainable machine learning model is proposed to tackle the problem of predictive analysis of chronic disease from EHR data. The performance of various machine learning techniques such as Logistic Regression, Random Forest, Support Vector Machine, XGBoost and Artificial Neural Networks is compared in predicting the disease. In order to enhance the interpretability, methods of Explainable Artificial Intelligence (XAI) are embedded to explain the model predictions in a patient-specific and global manner, such as SHAP and LIME. Through experimental results, we show that the proposed explainable models are both interpretable and have high predictive accuracy and strong ROC-AUC performance. The results demonstrate the clinical promise of explainable ML systems for providing healthcare decision-support applications that are transparent, reliable, and intelligent.