An Explainable Predictive Model for Diabetes Detection Using Shapley Additive Explanations Approach

The chronic and incurable disease of diabetes that is not only marked with increasing prevalence but is also linked to other life-threatening diseases like cardiovascular diseases, tissue injuries and retinopathy. With this the early detection and diagnosis of the disease becomes essential. This paper, proposes a novel explainable predictive model for diabetes detection using Shapley additive explanations approach which will not just predict the disease but also highlight the most contributing features contributing to our result. Through the utilization of PIDD dataset, diabetes has been predicted by using three machine learning classifiers i.e. Random Forest, XGBoost and Support Vector Machine. The proposed explainable model outperforms the existing studies by giving an accuracy of 82%. The results were then interpreted using Shapley Additive Explanations (SHAP) which further explained that the most contributing features throughout the study in each model were ‘Glucose’, ‘BMI’ and ‘Age’ among others. This explanation is meant to aid the medical community in their diagnosis and effective recommendations to the patients.