Predicting Diabetic Distress and Emotional Burden in Type-2 Diabetes Using Explainable AI

Diabetic distress is a significant psychological burden affecting many individuals with Type-2 Diabetes Mellitus. Despite its prevalence, it remains underrecognized, and its impact can be complex. This study explores the integration of multimodal data sources and machine learning to identify diabetic distress and emotional burden in patients with Type-2 Diabetes Mellitus. The study highlights the use of Explainable Artificial Intelligence to make predictions more transparent. It combines patient information like age, gender, laboratory results, and survey scores to train and compare machine learning methods. Models tested include Ridge regression, Lasso regression, linear and logistic regression, neural networks, support vector machines, random forests, and various boosting techniques. Among all of these, Extreme Gradient Boosting with SHapley Additive exPlanations delivered the best results. It was 96.14% accurate, 0.94 precise, had an ROC-AUC of 0.98, an F1-score of 0.95, and a recall of 0.95. The approach also generates explanations of why a specific prediction is being generated. For instance, it shows how specific blood glucose levels or distress subscale scores influence the predicted emotional load. These clear explanations enable clinicians to make sense of both the model’s performance and the reasoning behind it. Such transparency is critical to the building of trust in machine learning-based decision support systems for diabetes care takers influencing DDS.