Predicting Diabetic Patient Hospital Readmission Using Optimized Random Forest and Firefly Evolutionary Algorithm

Abstract

Diabetes is one of the most prevailing diseases worldwide. The number of hospitalized patients with diabetes is usually huge. Readmission in the hospital is expensive, and early prediction of diabetes patient’s hospital readmission can reduce the cost and help healthcare professionals evaluate the quality of healthcare services at the hospital. The proposed study aimed to develop an early prediction model for diabetes readmission and identify the significant factors that lead to readmission of diabetes patients. The early prediction will reduce the risk of hospital readmission. Several machine learning classifiers, such as Logistic Regression (LR), Decision Tree (DT), and Random Forest (RF), were applied. Firefly bio-inspired technique was used for feature selection and model optimization. Synthetic Minority Oversampling Technique (SMOTE) was applied to alleviate the data imbalance problem. The performance of the classifiers was compared using different feature sets. Experiments showed that RF outperformed the other models using reduced features selected by the Firefly algorithm. The study achieved the highest accuracy, precision, recall, and Area Under Curve (AUC) of 0.99, 0.99, 0.94, and 0.98, respectively. The results show the significance of the proposed model in diabetes readmission prediction. As a result, it is suggested that other system models and multiple data sets be investigated in order to obtain better results and identify significant features for early readmission prediction in diabetic patients.