Explainable neural networks in diabetes mellitus prediction

Abstract

Artificial Intelligence (AI) has been widely applied in healthcare for several purposes, especially in disease prediction enabling physicians to more accurately diagnose patients' conditions. Results generated by traditional AI models are difficult to justify due to the opaqueness of the models. Thus, making it difficult for physicians to trust the results and use them in real-life practice. Recent advancements in explainable AI (XAI) have made the results more reliable, making it possible for physicians to embrace AI in clinical practice. Explainable deep neural network (xDNN) is a machine learning technique that can enhance diabetes mellitus disease prediction and explain the results. This chapter focuses on using explainable neural networks in diabetes mellitus prediction. It provides valuable insights on key steps and techniques for diabetes mellitus prediction using explainable neural networks (xNNs). In particular, the sequence for implementing the model using R programming software was discussed. In order to demonstrate the implementation of xNNs in diabetes mellitus prediction, the Pima Indian diabetes mellitus datasets were used. The model was assessed based on accuracy, sensitivity, specificity, precision, recall, and F1 score. Additionally, the chapter discussed the different methods of implementing explainability in XAI's and provided a clear illustration using the variable importance tool in R. The results revealed the effect of each variable on the overall model. We found that the variable importance varies with the network architecture. Overall, diabetes pedigree functions are the least important predictor of diabetes mellitus in the model.