Levaraging Big Data Analytics for Enhanced Clinical Decision Making

Hospital readmissions among diabetic patients represent a critical challenge in modern healthcare, contributing to increased operational costs and reduced quality of patient care. Predicting readmissions is inherently difficult due to the complexity and high dimensionality of clinical data, compounded by severe class imbalance between readmission and non-readmission cases. This paper proposes an interpretable machine learning framework that integrates Synthetic Minority Oversampling Technique (SMOTE) for imbalance handling, ensemble learning models (Logistic Regression, Random Forest, and XGBoost) for prediction, and SHAP (SHapley Additive exPlanations) for model interpretability. Experimental results demonstrate that XGBoost achieves the highest ROC-AUC score of 0.668, outperforming Logistic Regression (0.633) and Random Forest (0.664). Key predictors identified through SHAP analysis include the number of inpatient visits, gender, discharge disposition, and diabetes medication status. The system is deployed using a Streamlit-based interactive dashboard for real-time clinical decision support, providing risk scores from 0 to 100 and categorizing patients into Low, Medium, and High risk bands. Leveraging Big Data Analytics frameworks discussed in recent literature, the proposed system effectively balances predictive performance and clinical interpretability, making it suitable for real-world healthcare applications