Predictive Modeling of COVID-19 Readmissions: Insights from Machine Learning and Deep Learning Approaches.

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Public Health

Background:

• COVID-19 readmissions pose a significant challenge to healthcare systems globally.
• Accurate prediction of readmission risk is crucial for effective resource allocation and patient management.
• Malaysia faces the challenge of managing COVID-19 patient care and preventing readmissions.

Purpose of the Study:

• To develop and evaluate artificial intelligence (AI) models for predicting COVID-19 readmission risk in Malaysia.
• To compare the performance of various machine learning (ML) and deep learning (DL) algorithms in classifying COVID-19 readmissions.
• To identify the most effective AI techniques for mitigating healthcare resource strain and improving patient outcomes.

Main Methods:

• Dataset description and pre-processing.
• Data balancing techniques including Random Oversampling (ROS), Borderline SMOTE (BSMOTE), and Adaptive Synthetic Sampling (ADASYN).
• Application and evaluation of nine ML and ten DL techniques using five-fold cross-validation, with hyperparameter optimization via Optuna.

Main Results:

• CatBoost demonstrated superior performance, achieving the highest accuracy (0.9882 ± 0.0020) and AUC (1.0000 ± 0.0000) with ROS.
• CatBoost maintained its leading performance across BSMOTE and ADASYN data balancing methods.
• Deep learning models like SAINT (with ROS) and TabNet (with BSMOTE and ADASYN) also showed strong results, alongside Decision Tree ensembles (Random Forest, XGBoost).

Conclusions:

• AI, particularly the CatBoost model, shows high efficacy in predicting COVID-19 readmission risk in Malaysia.
• The developed methodology provides valuable tools for healthcare providers to manage resources and enhance patient care.
• The study highlights the potential of advanced ML and DL techniques in addressing public health challenges posed by infectious diseases.