Predictive Analytics for Diabetic Patient Care: Leveraging AI to Forecast Readmission and Hospital Stays

Abstract

Predicting hospital readmission and length of stay (LOS) for diabetic patients is critical for improving healthcare quality, optimizing resource utilization, and reducing costs. This study leverages machine learning algorithms to predict 30-day readmission rates and LOS using a robust dataset comprising over 100,000 patient encounters from 130 hospitals collected over a decade. A comprehensive preprocessing pipeline, including feature selection, data transformation, and class balancing, was implemented to ensure data quality and enhance model performance. Exploratory analysis revealed key patterns, such as the influence of age and the number of diagnoses on readmission rates, guiding the development of predictive models. Rigorous validation strategies, including 5-fold cross-validation and hyperparameter tuning, were employed to ensure model reliability and generalizability. Among the models tested, the Random Forest algorithm demonstrated superior performance, achieving 96% accuracy for predicting readmissions and 87% for LOS prediction. These results underscore the potential of predictive analytics in diabetic patient care, enabling proactive interventions, better resource allocation, and improved clinical outcomes.