Explainable Ensemble-Based Machine Learning Models for Detecting the Presence of Cirrhosis in Hepatitis C Patients

This study aimed to develop an accurate, explainable machine learning (ML) model to detect liver cirrhosis in patients with hepatitis C. Given that cirrhosis is a severe, often asymptomatic complication of chronic hepatitis C, early detection is critical for effective clinical management. The authors sought to improve upon existing diagnostic methods, which are often invasive (biopsy) or limited by availability and accuracy.

The researchers utilized a dataset of 1,385 Egyptian patients from the UCI Machine Learning Repository. They pre-processed the data by binarizing the target variable (cirrhosis vs. non-cirrhosis), removing outliers using the interquartile range method, and addressing class imbalance via random oversampling. Four ensemble-based algorithms—Random Forest, Gradient Boosting Machine, Extreme Gradient Boosting, and Extra Trees—were evaluated using 10-fold cross-validation. The study also employed Sequential Feature Selection (SFS) to identify the most predictive attributes and used Explainable AI (XAI) techniques, specifically SHAP and LIME, to interpret the model's decision-making process.

The Extra Trees model outperformed the other tested algorithms, achieving an accuracy of 96.92%, a recall of 94.00%, and a precision of 99.81%. By using SFS, the researchers successfully reduced the feature set from 28 to 16 variables while maintaining high diagnostic performance. The integration of XAI tools provided transparency into the model's predictions, which is essential for building trust among medical professionals who may be skeptical of "black-box" diagnostic tools.

This research demonstrates that ensemble-based ML models can provide a non-invasive, highly accurate alternative for identifying cirrhosis in hepatitis C patients. By reducing the number of required clinical features and providing interpretable results, the proposed framework offers a practical tool that could assist clinicians in making timely treatment decisions, potentially reducing the need for more invasive diagnostic procedures.