Explainable AI Approach in Diabetes Disease Classification Using Gradient Boosting Algorithm

Abstract

The classification of diabetes has become a critical focus in the healthcare domain due to the disease's rising prevalence and its severe impact on global health. While machine learning methods, such as Gradient Boosting Algorithm (GBA), have shown exceptional performance in predicting diabetes, the interpretability of these models remains a challenge for practical implementation in clinical settings. This study introduces an Explainable AI (XAI) approach to enhance the transparency and interpretability of the Gradient Boosting Algorithm for diabetes classification. Using clinical indicators such as HbA1c levels, Body Mass Index (BMI), and other risk factors, the model achieves high classification accuracy while providing insights into the feature contributions through visualization techniques. SHAP (SHapley Additive exPlanations) was utilized for detailed global and local explanations, while LIME (Local Interpretable Model-agnostic Explanations) offered localized insights into individual predictions. Both LGBoost and XGBoost were compared on the same clinical dataset, where LGBoost achieved an accuracy of 97.27% and XGBoost slightly outperformed with an accuracy of 97.36%, suggesting its marginal advantage in this dataset. The results demonstrate the potential of integrating XAI in machine learning workflows to balance performance and interpretability, thereby fostering trust among healthcare practitioners and aiding in informed decision-making. This research contributes to advancing the application of explainable models in medical diagnostics