Statistical Analysis and Machine Learning Prediction of Diabetes Risk Using BRFSS Data

One of the most urgent global health concerns today is diabetes mellitus, which highlights the necessity of early detection of those who are more vulnerable in order to enhance preventative measures and disease management. In order to forecast the risk of diabetes, this study presents a thorough framework that combines explainable artificial intelligence (XAI), supervised machine learning models, and statistical hypothesis testing. The 2015 Behavioral Risk Factor Surveillance System (BRFSS) dataset, which comprises records from 253,680 persons and contains 21 variables pertaining to lifestyle choices, clinical health indicators, and demographics, is used to assess the suggested method. For a number of critical risk factors, such as body mass index (BMI), blood pressure, cholesterol, physical activity, and other relevant variables, statistical analyses verified the existence of substantial differences between diabetic and non- diabetic populations (p < 0.001). The Synthetic Minority Oversampling Technique (SMOTE paired with random undersampling) was used to handle class imbalance before developing the Logistic Regression, Random Forest, and XGBoost algorithms for predictive modeling. With an accuracy of 82.16% and a receiver operating characteristic area under the curve (ROC- AUC) of 0.8234, XGBoost outperformed the other tested models. The most significant factors influencing diabetes risk were high blood pressure, self- reported general health, raised cholesterol, BMI, and age, according to SHAP- based explainability, which was used to improve model transparency. In conclusion, the suggested framework effectively blends interpretability and significant predictive potential, making it ideal for widespread diabetes risk screening in public health settings at the population level.