Assessing advanced machine learning techniques for predicting hospital readmission

Abstract

Predicting the probability of hospital readmission is one of the most important healthcare problems for satisfactory, high-quality service in chronic diseases such as diabetes, in order to identify needful resources such as rooms, medical staff, beds, and specialists. Unfortunately, not many studies in the literature address this issue. Most studies involve forecasting the probability of diseases. For prediction, several machine learning methods can be implemented. Nonetheless, comparative studies that identify the most effective approaches for the method prediction are also insufficient. With this aim, our paper introduces a comparative study in the literature across five popular methods to predict the probability of hospital readmission in patients suffering from diabetes. The selected techniques include linear discriminant analysis, instance-based learning (K-nearest neighbors), and ensemble-based learning (random forest, AdaBoost, and gradient boosting) techniques. The study showed that the best performance was in random forest whereas the worst performance was shown by linear discriminant analysis.