Advanced Predictive Modeling of Diabetic Readmissions Using Machine Learning and Essential Health Metrics

Abstract

The rising prevalence of Type 2 Diabetes Mellitus (T2DM) has underscored the need for predictive models that can effectively assess readmission risks among diabetic patients, thereby supporting improved healthcare management and resource allocation. This study presents an advanced approach for predicting diabetic patient readmissions by incorporating critical health metrics—Body Mass Index (BMI), HbA1c, cholesterol (new-Chol), and triglycerides (new-TG)—and employing a variety of machine learning algorithms. Using a dataset of diabetic patients with readmission intervals between 2 and 30 days, we conducted extensive data preprocessing, feature analysis, and tested several classifiers, including Support Vector Classifier (SVC), Decision Tree (DT), K-Nearest Neighbors (KNN), Logistic Regression (LR), Random Forest (RF), AdaBoost, and Gradient Boosting (GB), with cross-validation and feature selection for accuracy optimization. The ensemble model, which combined SVC, Random Forest, and XGBoost, outperformed individual models, achieving the highest accuracy at 98.6%, demonstrating the efficacy of ensemble methods in enhancing predictive performance and robustness. The Gradient Boosting Classifier also performed well independently, achieving 97.6% accuracy, while demographic factors, especially age, were found to significantly impact readmission patterns, offering valuable insights for personalized diabetes management. This predictive model equips healthcare providers with a critical tool for early intervention and improved patient outcomes through targeted diabetes care strategies.