Diabetes Risk Factors Examination using Interpretable Light Gradient Boosting Machine

The high prevalence of diabetes mellitus, along with its asymptomatic onset in the early stages, poses significant challenges in the field of public health. Data-driven exploratory studies on risk factors are essential to effectively identify them to support early intervention and prevention strategies. This study aims to identify factors influencing the risk of diabetes by employing an interpretable LightGBM model to estimate diabetes risk, supplemented by SHAP value–based analysis for model interpretation. However, one critical issue lies in the availability of comprehensive, population-level datat that not only captures clinical indicators but also behavioral and socioeconomic determinants of health. Therefore, we deploy our model on the Behavioral Risk Factor Surveillance System (BRFSS) data. Based on model evaluation, we achieved performance metrics with an accuracy of (76.60±5.35)%, G-mean of (76.58±5.35)%, and ROC-AUC of (84.24±3.88)%. General health status, blood pressure, and BMI should receive particular attention in diabetes prevention efforts due to the high SHAP value. This study is expected to provide deeper insights into the risk factors associated with diabetes mellitus.