Date of Publication :15th March 2025

Abstract: Hospital readmission is one of the significant health- care challenges because it often indicates a gap in patient management and transitions of care. Accurate prediction of readmission likelihood can enhance healthcare delivery, reduce costs, and improve outcomes for patients. This article explores the application of machine learning techniques—Logistic Regression, Decision Tree, and Random Forest—to predict hospital readmis- sions using a more comprehensive dataset of patient records. The dataset was preprocessed by the elimination of missing values and encoding categorical variables into numerical values, along with the removal of irrelevant features. Class imbalance was carried out using Synthetic Minority Over-sampling Technique (SMOTE) to ensure excellent generalization of the model. The performances of the models are assessed in accuracy, pre - cision, recall, F1-score, and AUC-ROC. Baseline model: logistic regression. Interpretation into which factors are most impactful on readmission regarding the time spent in the hospital and number of medications. Decision Tree and Random Forest utilize a non-linear relationship towards improvement of the prediction accuracy of the models. The best accuracy among these models is of Random Forest, along with the best trade-off between precision and recall. Meanwhile, Logistic Regression became to be a very interpretable one. This research puts the interest of using machine learning for fighting against healthcare issues, brought forward as this paper will present a data-driven approach that predicts and mitigates hospital readmission. Those models help in identifying high -risk patients, thereby aiding the healthcare providers in targeted interventions toward optimum patient care and resource utilization.

Reference :