Type ۲ Diabetes Risk Prediction Using Artificial Intelligence Models: A Comparative Analysis of XGBoost, DNN, and Random Forest

Type ۲ diabetes (T۲D) is an increasingly significant public health challenge, necessitating accurate and reliable predictive tools for early intervention. This study aimed to evaluate and compare the performance of three artificial intelligence models—XGBoost, deep neural networks (DNN), and Random Forest—in predicting the risk of T۲D. Data were collected from three major sources—dietary patterns, key biomarkers (HbA۱c, fasting glucose, insulin, blood lipids), and demographic and lifestyle characteristics—and analyzed after normalization, feature selection, and missing data imputation. The models were trained and tested using an ۸۰/۲۰ split and assessed through five-fold cross-validation. Results showed that the XGBoost model achieved the highest accuracy (۹۲.۵%) and AUC-ROC (۰.۹۶), with only a ۰.۵% deviation from actual data. Feature importance analysis indicated that HbA۱c (۲۱%), BMI (۱۷%), and sugar intake (۱۲%) were the most significant predictors. The DNN model demonstrated acceptable performance in capturing nonlinear patterns but had limited interpretability. Random Forest showed lower accuracy compared to XGBoost. These findings suggest that XGBoost can serve as an effective tool for early screening and for designing personalized dietary and lifestyle interventions. The development of explainable models and prospective studies is recommended to enhance generalizability.

کلیدواژه ها:

Type ۲ diabetes prediction ، machine learning algorithms ، artificial intelligence models