Science Direct

Hospital readmissions increase the healthcare costs and negatively influence hospitals’ reputation. Predicting readmissions in early stages allows prompting great attention to patients with high risk of readmission, which leverages the healthcare system and saves healthcare expenditures. Machine learning helps in providing more accurate predictions than current practices. In this work, an approach that balances between data engineering and neural networks’ ability to learning representations is proposed for predicting hospital readmission among diabetic patients. A combination of Convolutional neural networks and data engineering were found to outperform other machine learning algorithms when employed and evaluated against real life data.