Follow-up Period Classification of Type 2 Diabetes Patients using Data Mining Techniques

Abstract

Objective: This study investigates the use of high-performance data mining techniques to predict the follow-up period of diabetes patients. Material and Methods: The diabetes dataset was obtained from Pak Phanang hospital in Nakhon Si Thammarat, Thailand. The hospital acquired the data between January 1 and December 31, 2022. The hospital-based retrospective study was based on 2,042 records, featuring 14 independent factors; including age, gender, systolic blood pressure, diastolic blood pressure, body mass index, pulse, weight, height, waist, smoking, drinking, parental history of diabetes, and fasting blood sugar and creatinine levels. To predict the follow-up period of diabetes patients, six well-known classification models were employed: Random Forest (RF), Extra Trees Classifier (ETC), Adaptive Boosting (Adaboost), Support Vector Machine (SVM), K-Nearest Neighbor (KNN), and Artificial Neural Network (ANN). Class imbalances were addressed by using the Synthetic Minority Oversampling Technique (SMOTE), and feature importance was handled using the RF model. Results: The experimental results demonstrated that, by applying SMOTE together with Random forest feature selection, the Support vector machine outperformed the other models; exhibiting the highest performances with a weighted precision of 0.9296. Conclusion: The results indicated that incorporating both SMOTE and feature selection resulted in significantly improved accuracy in predicting the follow-up period of diabetes patients for most models. Therefore, doctors and related healthcare providers could employ our proposed web-based tool to effectively schedule follow-up care for diabetes patients.