A Machine Learning application of using different models for Predictive Modeling for Early Hospital Readmission in Diabetic Patients within 30 Days of Discharge.

This project addresses the critical issue of early hospital readmission among diabetic patients by utilizing a decade's worth of clinical data (1999-2008) from 130 US hospitals and integrated delivery networks. The study harnesses advanced data mining techniques to explore and preprocess the dataset, focusing on identifying key factors that influence readmission rates. In addition to the proposed initial models Decision Tree, K-NN, and SVM, I have expanded my study to include Logistic Regression, Naive-Bayes, and Random Forests to enhance predictive accuracy and robustness.

The models were evaluated based on their performance in predicting early readmissions, with a focus on precision, recall, F1-score, and notably, the area under curve (AUC). The comparative analysis revealed that Optimized Random Forest not only excelled in standard performance metrics but also demonstrated superior AUC, making it the model of choice. These findings highlight the efficacy of machine learning in facilitating healthcare providers to formulate proactive interventions and personalized care, significantly enhancing patient outcomes and mitigating healthcare costs.