Predicting diabetes using Machine Learning and several eXplainable Artificial Intelligence approaches

Diabetes is a biochemical disorder defined by long-term and chronic elevation of blood sugar levels, known as hyperglycemia. The symptoms encompass frequent urination, dehydration, increased appetite, and weight swings. Generally, diabetes prediction can be determined using either a glucometer or an A1C blood sugar test. Diabetes has become increasingly challenging to treat and, as a result, has become a disease that poses a significant risk to life. Presently, artificial intelligence and machine learning (AI/ML) were widely utilized for detection and management of such illnesses. In this study, we employed artificial intelligence AI and ML approaches to diagnose diabetes using clinical markers. Among the eight machine learning models employed, the random forest model and lightgbm produced the most favorable results, attaining an accuracy of 0.77. Explainable artificial intelligence alludes to the ability of an AI system for offering lucid and comprehensible explanations for its choices and behaviors. This research utilizes XAI methods like SHAP, ELI5, Qlattice, as well as LIME to guarantee interpretable and transparent results. This model facilitates the detection of individuals with a high-risk characteristic, enabling the implementation of early detection of diabetes.