A cross dataset meta-model for hepatitis C detection using multi-dimensional pre-clustering

Area of Science:

• Hepatology
• Medical Informatics
• Machine Learning in Healthcare

Background:

• Hepatitis C virus (HCV) infection causes liver inflammation, potentially leading to chronic liver disease, hepatocellular carcinoma (HCC), and death.
• Early diagnosis and treatment of HCV are critical to prevent severe health complications.
• Standard biochemistry tests provide valuable data for patient assessment.

Purpose of the Study:

• To develop and evaluate a machine learning model for precise identification of hepatitis C (HCV) using standard biochemistry test parameters.
• To enhance diagnostic accuracy by employing a novel pre-clustering approach and a stacked meta-model.
• To compare the performance of the developed meta-model against established baseline machine learning models.

Main Methods:

• A hybrid dataset was created by merging two common HCV datasets, with a hold-out set for validation.
• A multi-dimensional pre-clustering approach (k-means and k-modes) was utilized for feature extraction.
• A stacked meta-model was trained using the enhanced dataset, incorporating XGBoost, K-nearest neighbor, support vector classifier, and random forest (RF) algorithms.

Main Results:

• The stacked meta-model achieved a diagnostic accuracy of 94.82%.
• This performance surpassed the baseline random forest (RF) model, which scored 94.25%.
• Explainable artificial intelligence (XAI) techniques were employed to interpret model predictions.

Conclusions:

• Machine learning, particularly the developed stacked meta-model, shows significant promise for accurate and efficient hepatitis C (HCV) diagnosis.
• The integration of pre-clustering for feature engineering enhances diagnostic capabilities.
• This approach offers a valuable tool for early HCV detection, potentially improving patient outcomes.