Explainable AI-Based Diabetes Mellitus Risk Prediction Using Naive Bayes and Support Vector Machine

Abstract

Diabetes Mellitus (DM) is a chronic metabolic disease that continues to increase globally and requires effective early detection to reduce the risk of serious complications. Machine learning has been widely adopted as an approach for predicting diabetes risk; however, most existing models are still black-box in nature, making them difficult to interpret and less useful for clinical decision-making. In addition, the problem of class imbalance in medical datasets often causes models to be biased toward the majority class, reducing their sensitivity in detecting high-risk patients. This study aims to develop and compare diabetes risk prediction models using the Naive Bayes and Support Vector Machine (SVM) algorithms with an Explainable Artificial Intelligence (XAI) approach. Class imbalance was addressed using the Synthetic Minority Over-sampling Technique (SMOTE) applied to the training data. Interpretability was analyzed using SHAP (SHapley Additive exPlanations) for global feature importance and LIME (Local Interpretable Model-agnostic Explanations) for local instance-level explanations. The dataset used was the Diabetes Health Indicators Dataset from the BRFSS 2015 survey, publicly available on Kaggle, with a sample of 50,000 records and 22 variables. Evaluation results showed that SVM achieved an accuracy of 84.17%, while Naive Bayes achieved a higher recall of 77.76%, indicating better sensitivity in detecting diabetes cases. SHAP analysis identified GenHlth, HighBP, BMI, HighChol, and Age as the most influential risk factors globally, while LIME provided individual-level explanations. This research contributes a prediction model that is not only accurate but also transparent and clinically interpretable.