Diagnosis of Hepatitis C Patients via Machine Learning Approach: XGBoost and Isolation Forest

Abstract

Although the transmission of Hepatitis C through blood transfusion is getting less and less prevalent with the use of anti-HCV tests for blood donors, availability and practice of screening remain low in developing countries. This results in Hepatitis C patients who are unaware of their condition until it worsens to become chronic liver diseases that are diagnosed through more costly or invasive methods–liver biopsy and radiology scans. Due to these limitations of the current methods of diagnosis, this study seeks to develop a machine learning model to diagnose patients with different stages of liver disease: hepatitis c, liver fibrosis, and cirrhosis. In this research, machine learning algorithms were applied to a dataset containing HCV patient information, and the algorithms were evaluated for their accuracy and performance in classifying the patients with the proper diagnosis. Findings from the study indicated that XGBoost can most accurately classify patients with an accuracy score of 95.48, but other algorithms used had high accuracy scores as well: the algorithm with the lowest accuracy score–Decision Tree–still had a score of 92.66. The second experiment also showed that the Isolation Forest algorithm could detect and isolate the suspect blood donors of the data with a relatively high accuracy of 93.22%. As both experiments of the study yielded a machine learning model of high accuracy, the algorithms used can be implemented into a diagnostic kit for liver disease to be used in developing countries where accessibility to current diagnosis tools is limited.