Risk Prediction of Heart Disease using Deep SHAP Techniques

The automatic identification o f h eart d iseases is a challenging problem in the medical field, a nd i t constitutes the primary cause of death. Heart failure prediction, one of the symptoms of cardiovascular disease, has gained importance among physicians. In addition to increasing feature ranking and clinical prediction, deep learning also helps with output interpretation for medical practitioners. Therefore, the idea behind Explainable Artificial Intelligence (XAI) is to address the problem of deep learning algorithms in the healthcare domain not being explainable, in this case, and give the user knowledge about the model's inner working process in an understandable format. In our proposed work, convolutional neural networks are used to predict the disease, and the Deep SHAP model is used to visualize the model predictions with an accuracy of 0.90, sensitivity of 0.97, and F1-score of 0.86 for class 1. The results section includes metrics computed for two classes, such as recall, precision, and F1-score.