Intelligent Models for the Detection of Liver Cirrhosis with Emphasis on Imbalanced Classes

Abstract

Liver cirrhosis is a leading cause of morbidity and mortality worldwide, with an increasing prevalence associated with multiple etiologies. Accurate prediction of survival in cirrhotic patients is crucial for risk stratification and the optimization of therapeutic resources, particularly in identifying candidates for liver transplantation. This study comparatively evaluated three machine learning approaches: Random Forest with class weights, Artificial Neural Network with SMOTE oversampling, and a Fuzzy Logic classifier with reinforced rules, using the public dataset from the Mayo Clinic Trial (n=8,181). The class distribution showed extreme imbalance: 62.5% censored, 33.9% deceased, and 3.6% transplanted. The results showed that Random Forest achieved the best overall performance (Balanced Accuracy=0.652, F1-macro=0.666), with particularly outstanding accuracy in the majority classes (F1-censored=0.866, F1-death=0.760).