Predicting 30-Day Hospital Readmission in Diabetic Patients Using Machine Learning

Hospital readmissions are a significant problem in healthcare, particularly in diabetic patients. The research question guiding this study was: Can 30-day hospital readmission for diabetic patients be predicted using common machine learning models applied to hospital record data? This study aims to predict this using three machine learning models: Logistic Regression, Random Forest, and XGBoost. The analysis was conducted on the Diabetes 130-US Hospitals dataset from the UCI Machine Learning Repository, which contains over 100,000 patient records. After data cleaning, 98,053 complete records with 20 features were used for model training and evaluation. Random Forest achieved the highest accuracy at 88.73%, Logistic Regression at 88.67%, and XGBoost at 88.59%. Moreover, the XGBoost model’s Area Under the Curve (AUC) was only 0.653, indicating performance slightly better than random guessing. The number of lab procedures were found to be the strongest predictor of readmission, followed by different diagnoses and number of medications. The models struggled to identify readmitted patients due to dataset imbalance, with only 11.3% of patients readmitted. These findings suggest that while machine learning can predict readmission status with reasonable overall accuracy, current models are not reliable enough for clinical use without additional features and improved data quality.