HCV Ensemble Learning: A Reproducible Machine Learning Study for Hepatitis C Prediction

This study conducts a reproducible research based on the UCI HCV dataset, comparing Multilayer Perceptron (MLP), Bayesian networks, QUEST decision trees, and ensemble learning methods. The results show that ensemble learning performs excellently in hepatitis C prediction. The study provides complete code and analysis documents, offering references for reproducible research in the medical AI field and the development of auxiliary diagnostic tools.

Approximately 58 million people worldwide are chronically infected with Hepatitis C Virus (HCV), and 290,000 people die from related complications each year, making early diagnosis crucial. Traditional diagnosis relies on blood tests and doctors' experience, which has limitations when dealing with multi-dimensional indicators. This study uses the UCI HCV dataset (615 records, 13 biochemical indicators + 1 target variable) to explore the potential of machine learning for auxiliary diagnosis.

The study uses four methods: 1. Multilayer Perceptron (MLP): A fully connected neural network that captures nonlinear interactions; 2. Bayesian Network: A probabilistic graphical model that handles uncertainty; 3. QUEST Decision Tree: A fast and unbiased statistical decision tree; 4. Ensemble Learning: Combines the prediction results of the above three models (majority/weighted voting) to improve performance using model complementarity.

Comparison of reproduction results with the original paper: MLP (94.15% vs 94.10%), Bayesian Network (94.47% consistent), QUEST (94.63% consistent), and the ensemble model's reproduction accuracy reached 99.32% (original paper:95.59%). The reason for the difference lies in the different implementation details of Bayesian networks between SPSS and sklearn, leading to changes in the voting divergence pattern and improving the ensemble effect.

Project files include the dataset, Excel analysis result tables, and Python reproduction scripts (requires installation of libraries like scikit-learn to run). Feature importance analysis based on QUEST shows that specific enzyme indicators and protein levels are key factors for HCV prediction, providing support for clinical testing priorities.

Methodological contributions: Demonstrates the migration path from SPSS to Python and verifies the effectiveness of ensemble strategies; Clinical applications: Auxiliary diagnosis, early screening, resource optimization; Limitations: Small sample size (615), single data source, need for clinical validation and regulatory approval.

This study proves the high accuracy of ensemble learning in HCV prediction (99.32% reproduction result), providing direction for the development of medical AI auxiliary diagnostic tools. As a reproducible case, this project has reference value for medical AI researchers and data scientists.