Predicting hospital readmissions in diabetes patients: A comparative study of machine learning models

This study addresses the high hospital readmission rates among diabetes patients, which contribute to increased healthcare costs and strain on resources. By leveraging machine learning (ML) techniques, the objective is to predict readmissions and help healthcare providers identify high-risk patients for early intervention. Six machine learning models—Logistic Regression, Random Forest, Gradient Boosting, XGBoost, LightGBM, and CATBoost—were employed using the Diabetes 130-US hospitals dataset, incorporating patient demographics, clinical data, and discharge information. The models were evaluated based on metrics such as accuracy, precision, recall, and AUC-ROC. Among the models, CATBoost performed the best, achieving an AUC score of 0.70 and an accuracy of 64.2%. The most critical predictive features were the number of inpatient visits, medications prescribed, and the length of hospital stays. These results highlight the potential of machine learning in predicting hospital readmissions, providing actionable insights for improving patient outcomes. Future research should explore integrating real-time health data from wearables and examine the role of social determinants to further enhance predictive accuracy and optimize healthcare resources.