Leveraging XAI for Discovering Crucial Demographic, Clinical and Pathological Diabetes Mellitus Biomarkers

Diabetes Mellitus (DM) is a chronic metabolic disorder [1] which results in hyperglycemia caused by defects in insulin secretion or action. Prevalence of DM especially Type 2 Diabetes Mellitus(T2DM) is steadily increasing worldwide, making it a daunting health challenge. Multiple factors such as sedentary lifestyles, family history, genetics, unhealthy diets, obesity, and ageing drive its occurrence. Effective management of diabetes requires early detection, accurate diagnosis, and personalized treatment strategies. Biomarkers, quantifiable indicators of biological processes and disease states, play a crucial role in disease diagnosis, prognosis, and treatment monitoring. Unlocking significant biomarkers for DM detection is complex due to its multifactorial nature and intricate interactions. In recent years, Explainable Artificial Intelligence (XAI) has emerged as a promising approach for modelling and analyzing biomarkers, offering interpretability through feature importance techniques. This paper highlights the potential of XAI in early detection of T2DM and its progression. We delved deep into three publicly available diabetes datasets viz PIMA, Mendeley and Vanderbilt to get insights about top-tier clinical, pathological and demographic biomarkers essential for the timely detection of T2DM. We applied nine Machine Learning algorithms to determine the most accurate approach. XAI techniques of Shapley values, Permutation Importance and Decision Trees were utilized for comprehensive global interpretations of results and biomarker discovery. This study can be leveraged for clinical decision support systems to identify which biomarkers warrant regular monitoring for timely therapeutic intervention, thereby preventing progression of T2DM and its complications.