Explainable Artificial Intelligence-Based Diagnosis Assistant of Hepatitis C Virus

Abstract

Hepatitis C is a liver infection prevalent in developing countries, and early detection of this disease would significantly reduce the mortality rate. Advances in artificial intelligence have led to the development of medical diagnostics systems. However, the decisions gotten from these systems are not easily explainable to the end users. Data preprocessing, including feature scaling and oversampling using Synthetic Minority Oversampling Technique, was carried out on HCV data. Seven classifiers—logistic regression, decision tree, random forest, support vector machine, gradient boosting, K-nearest neighbor, and multilayer perceptron (MLP)—were implemented. The models were evaluated, and Shapley Additive Explanations (SHAP) values were employed for model interpretability. MLP with standard scaling has the best performance with an accuracy of 0.97 and a sensitivity and specificity of 1.00. The features with the most influence on the outcome are the albumin test, alkaline phosphatase, alanine transaminase, and aspartate aminotransferase, while sex and cholesterol had the least influence. A web-based diagnosis assistant was deployed for early diagnosis.