Hybrid intelligent systems for liver disease prediction: a demographic-aware machine learning framework

Authors:

Abstract
Background: Liver disease remains a major global health burden, often progressing undetected until advanced stages. Traditional diagnostic approaches, while accurate, are invasive, costly, and limited in accessibility. Objective: To address these challenges, we propose a hybrid intelligent framework that integrates demographic segmentation with advanced machine learning for the early detection of liver disease. Results: Two datasets were employed, including the Indian Liver Patient Dataset (ILPD, n = 583) and a large-scale dataset ( n = 29,787). Patients were stratified by age and gender into six groups, enabling segment-specific model development. Sixteen algorithms, including Random Forest, Support Vector Machines, XGBoost, and LightGBM, were evaluated using recursive feature elimination, resampling techniques, and Bayesian hyperparameter optimization. Segment-specific best models were integrated into a hybrid system through dynamic selection and ensemble strategies. The framework achieved 94.2% accuracy on ILPD and 99.8% on the large dataset, with consistent improvements across demographic groups. Feature analysis revealed distinct biomarker importance by age and gender, underscoring the need for tailored diagnostic approaches.. Conclusion: By combining demographic awareness, hybrid learning, and interpretability, this study offers a scalable, non-invasive, and clinically relevant tool for early detection of liver diseases, advancing personalized and accessible healthcare.

This publication has 35 references indexed in Scilit: