Towards an explainable machine learning model to reduce readmission risks for diabetes patients

Hospital readmission of Diabetes patients is a persistent burden on the healthcare industry. Artificial Intelligence (AI) based Machine Learning (ML) techniques offer the potential to predict readmission rates and related risk features for diabetic patients. However, complex machine learning-based solutions are often not explainable and hard to understand for the relevant parties. To this end, this study designs and implements an explainable model to predict readmission rates and identify the risk factors associated with readmission in patients with diabetes. The model employs various explainable visualization techniques, including the permutation importance plot, partial dependence plot (PDP), SHapley Additive exPlanations (SHAP), and interpretable classifiers on a publicly available dataset from US hospitals. The bagging random forest model yields the best results, achieving 89% accuracy and 67% precision. The explainability visualization techniques reveal that the number of inpatient admissions and emergency visits in a year is the two most critical risk factors for the readmission rate of diabetic patients. • It shows an explainable ML model to predict readmission risks for diabetic patients. • It addresses the class imbalance problem in the dataset and improves accuracy. • The Bagging Random Forest model achieved 89% accuracy and 67% precision. • Key risk factors include the number of inpatient admissions and emergency visits. • Various explainable AI techniques were used to improve model interpretability.