Towards Explainable Machine Learning in Early Diabetes Prediction

Abstract

This study explores the application of explainable machine learning techniques in the early prediction of diabetes, aiming to bridge the gap between complex model predictions and their interpretability in clinical settings. Utilizing the Pima Indian Diabetes Dataset, we preprocess data by addressing missing values, normalizing features, and selecting clinically relevant variables. Interpretable models such as decision trees, logistic regression, and rule-based classifiers are employed to provide transparent insights into diabetes risk factors. Techniques like SHAP values and LIME are integrated to further elucidate individual predictions, enhancing the clarity and trustworthiness of the models. The results indicate that high glucose levels and BMI are significant predictors of diabetes, aligning with clinical expectations. Through case studies, we demonstrate how these explainable models support informed clinical decisions, offering a robust framework for early diabetes detection and personalized patient care. This research underscores the importance of interpretability in machine learning, promoting its adoption in healthcare for better patient outcomes and more reliable diagnostic tools.