An Optimized XGBoost approach for Predicting Progression of Hepatitis C using Hyperparameter Tuning and Feature Interaction Constraint

Hepatitis C is a prominent viral infection affecting the liver, which has an estimated 58 million active cases throughout the world, with 1.5 million new infections adding to this tally every year. With a death toll of 290,000 by 2019, early diagnosis is the key to curing this infection with a recovery rate of 95%. In this paper, we aim to introduce an optimized XGBoost algorithm for predicting early progression to Hepatitis C which has been implemented on a labelled dataset with subjects of both genders and has been obtained from the UCI Machine Learning Repository. It contains data from 615 subjects aged between 19 and 77. The classification has been conducted primarily based on parameters such as Albumin (ALB), Alkaline Phosphate Enzyme (AKP), alanine aminotransferase (ALT), Bilirubin (BLT), Aspartate aminotransferase enzyme (AST), Cholesterol (CHOL), Serum cholinesterase (CHE) and gamma-glutamyl transferase (GGT).Our proposed methodology predicts progression towards Hepatitis C using XGBoost optimized using Monotonic Constraints and feature Interaction constraints. The prediction architecture is further optimized using regularization parameters and introducing boundary conditions such as Max- Depth, Learning rate and N-estimation. The resultant methodology predicts progression to Hepatitis C with an accuracy of 98.6%, which is higher when compared to other algorithms including Logistic Regression, LGBM, Decision Tree and SVM-RBF.