Hybrid Feature–Noise Adaptive Preprocessing for Liver Disease Prediction Using Statistical Learning Models

Abstract

Liver diseases pose a global health challenge, requiring accurate and reliable prediction methods for early diagnosis. This study introduces a Hybrid Feature–Noise Adaptive Preprocessing (HFNAP) algorithm that enhances data quality through adaptive imputation, dynamic outlier rejection, and correlation-based feature pruning. Using the UCI Hepatitis C Virus dataset, HFNAP statistically stabilizes data distributions and mitigates class imbalance while maintaining computational efficiency. Comparative evaluation with two contemporary methods—LPDS and Poly-SHAP—demonstrates that HFNAP achieves superior reduction in data skewness, feature redundancy, and imbalance ratio. The approach offers a transparent, statistically grounded preprocessing framework suitable for improving the performance and reliability of downstream classification models. Overall, HFNAP establishes a reproducible foundation for medical data preparation, bridging the gap between efficiency and interpretability in healthcare machine learning systems.