Enhanced Diabetes Prediction Using Salivary Enzymes and Machine Learning Techniques: A Comprehensive and Explainable Approach

Diabetes is a significant cause of death globally. People with diabetes are more susceptible to infections, possibly due to problems with their immune system. Diabetes mellitus (DM) is a disorder characterised by insufficient insulin production or ineffective utilisation of insulin, leading to elevated blood glucose levels. Timely identification of diabetes is essential for efficient control and therapy. Conventional techniques depend on intrusive blood testing, however recent advancements indicate that salivary enzymes can be used as non-invasive indicators. Using saliva for diagnosis is promising because collecting saliva is simple, doesn’t require invasive procedures, and is less technically demanding than drawing blood. Saliva is a special liquid that plays a crucial role in the mouth’s normal functioning. Detecting diabetes through blood tests can be challenging in certain groups like children, older adults, and those who are sick or weakened. This study introduces an improved framework for predicting diabetes by combining salivary enzymes with advanced machine learning algorithms and ensemble learning. To improve the model’s comprehensibility, we employed the Local Interpretable Model-agnostic Explanations (LIME) technique, which is a component of explainable AI (XAI). Multiple machine learning models are assessed using the Diabetes dataset, which comprises data from 300 patients. Simulation results reveal that XGBoost attains the maximum accuracy of 89% and elucidates the causes and variables contributing to diabetes, thereby assisting medical practitioners in decision-making for clinical diagnosis and treatment strategies.