A COMPREHENSIVE SURVEY OF PREDICTIVE MODELING TECHNIQUES FOR HEART FAILURE HOSPITAL READMISSION: CHALLENGES, RESEARCH GAPS, AND A PROPOSED THREE-PHASE RESEARCH FRAMEWORK
DOI: https://doi.org/10.65725/JCISE/2/2/004
JOURNAL OF COMPUTATIONAL INTELLIGENCE SCIENCE AND ENGINEERING (JCISE)
ISSN: 3107-8168Volume 2 Issue 2, Apr-Jun 2026

Abstract: 
Hospital readmission among heart failure (HF) patients remains a critical clinical and economic challenge globally. Despite advances in electronic health records (EHR) and the growing application of machine learning techniques, accurately predicting early readmissions for HF patients remains complex. This complexity are arises from the multifactorial nature of HF as a condition and the heterogeneity across patient populations. This survey synthesizes over fifteen key studies on state-of-the-art predictive modeling approaches for HF readmission. Techniques reviewed range from traditional logistic regression models to more advanced ensemble methods and deep learning architectures. The review critically analyzes dataset characteristics such as size and feature types, model architectures, feature engineering strategies aimed at enhancing predictive accuracy, methods for improving model interpretability, and challenges related to integrating predictive models into clinical workflows. The analysis reveals persistent research gaps, most notably in managing imbalanced datasets where readmission cases are underrepresented, incorporating unstructured EHR data such as clinical notes, and validating prediction models across diverse patient cohorts to ensure generalizability. To address these gaps, a novel three-phase research framework is proposed. First, focuses on exploratory data analysis and baseline modeling using freely accessible datasets to understand risk patterns and establish initial accuracy benchmarks. Additionally involves advanced machine learning development, utilizing sophisticated techniques like ensemble learning, deep feature engineering, and interpretability methods such as SHAP values to build robust and explainable models. Finally emphasizes rigorous validation using independent datasets and clinical utility assessment, including cross-dataset validation, clinical workflow integration, and simulation of potential impacts on reducing readmission rates. This phased approach balances methodological rigor with translational potential, placing strong emphasis on clinician engagement and ethical considerations. The survey’s findings aim to guide future research towards developing more robust, interpretable, and clinically deployable models for HF readmission prediction, ultimately improving patient outcomes and reducing healthcare burdens.

Authors: Mrs. Revatisivakumar, Dr. R. Preethi

Keywords: Heart failure, hospital readmission, machine learning, predictive modeling, electronic health records.