A Comparative Evaluation for Forecasting Diabetic Patients' Hospital Readmission within 30 Days through Machine Learning Techniques

Readmission to the hospital can place a heavy burden on patients and healthcare institutions with limited resources. Early intervention and better healthcare service quality can be achieved by identifying which patients are most likely to require readmission. The prevalence of diabetes among hospitalized patients is significantly increasing and so predictive analytics through machine learning methods to forecast hospital readmission, particularly for diabetic patients can be an excellent solution in this circumstance. Therefore, in this research, the electronic health record data from a hospital system has been used to predict readmissions of diabetic patients within 30 days of discharge using machine learning classification algorithms. The study has analyzed the significance o f various a ttributes, including commodities, demographic data, and prior hospitalization experience, with extensive data analysis and visualization. Several classification models 0 f machine1 earning, such a s gradient boosting, adaptive boosting, decision trees, random forests, K-Nearest Neighbours and logistic regression, have been trained and tested to make predictions. A comparative performance analysis with different types of parameters has been conducted, which reveals that the Adaboost classification technique provides the best accuracy with 88.3 % accuracy for predicting the readmission of diabetic patients.