A Diabetes Prediction Model with Visualized Explainable Artificial Intelligence (XAI) Technology

Diabetes is a group of non-communicable diseases (NCD) that cannot be cured by current medical technologies and can lead to various serious complications. Significantly reducing the severity of diabetes and its associated risk factors relies on accurate early prediction. Some machine learning algorithms have been developed to assist in predicting diabetes, but their predictions are not always accurate and often lack interpretability. Therefore, further efforts are required to improve these algorithms to achieve the level of clinical application. The aim of this paper is to find a high-performance and interpretable diabetes prediction model. Firstly, the dataset is subjected to necessary preprocessing, including missing value imputation using K-nearest neighbors (KNN) and data balancing using adaptive synthetic sampling (ADASYN). Then, with 10-fold cross validation, the predictive performance of six machine learning algorithms is compared in terms of accuracy, precision, recall, and F1 score. Finally, the prediction results are globally and locally explained using SHapley Additive exPlanations (SHAP) and Local Interpretable Model-agnostic Explanations (LIME). The experimental results demonstrate that the eXtreme Gradient Boosting (XGBoost) algorithm provides the best predictive performance. The visualized eXplainable Artificial Intelligence (XAI) techniques offer valuable explanatory information, helping healthcare professionals and patients better understand the risk and prediction results of diabetes.