Towards Interpretable Diabetes Prediction: A Comparative Study of Machine Learning Models with Explainable AI using LIME and SHAP

Diabetes is an ongoing condition that could lead to a global health crisis. Diabetes should not be disregarded as, if left untreated, it can lead to serious problems for a person, such as issues with the heart, kidneys, and blood pressure, and harm to the eyes, as well as other bodily organs. Diabetes is treatable if detected early on. In this study, we will estimate diabetes early on for more precision in a patient's or human body through utilizing a range of machine learning approaches like Logistic Regression, Decision Tree, Random Forest, XGBoost, CatBoost, Gradient Boost, Light GBM, Stacking, Soft voting, ANN and MLP. We used two datasets to apply machine learning models for this task. Compared to all other machine-learning models, the CatBoost classifier performs better in both datasets, with 83% and 98% accuracy. LIME and SHAP are two techniques from Explainable Artificial Intelligence (XAI). These methods are essential for deciphering and illustrating the CatBoost model's complex decision-making process. The knowledge gathered highlights the model's resilience and offers insightful details on the decision-making procedures, greatly advancing the field of diabetes prediction. To find out how effectively the CatBoost classifier worked with the dataset, we employed LIME and SHAP.