Machine Learning Algorithms for Survival Classification in Hepatocellular Carcinoma: A Comparative Study

Abstract

This study conducted a comprehensive comparison of three machine learning algorithms,
Random Forest (RF), Support Vector Machine (SVM), and XGBoost, for predicting survival
outcomes in hepatocellular carcinoma (HCC) patients using clinical data from 204 individuals.
The SVM model demonstrated superior performance with an accuracy of 80% (95% CI: 76–
84%), sensitivity of 80%, and specificity of 80%, outperforming both RF (accuracy: 78.3%,
95% CI: 74–82%; sensitivity: 76.7%; specificity: 80%) and XGBoost (accuracy: 76.7%, 95%
CI: 72–81%; sensitivity: 80%; specificity: 73.3%). All models showed strong discriminative
ability, with the area under the receiver operating characteristic curve (AUC-ROC) ranging
from 0.76 to 0.82, confirming their reliability in binary survival classification. Feature
importance analysis revealed that liver function markers (INR, albumin) and tumour
characteristics (size, AFP levels) were consistently ranked as top predictors across all
algorithms. The SVM's radial kernel effectively handled non-linear relationships in the data,
while RF provided valuable interpretability through its feature importance plots. These results
not only validate SVM as the most robust classifier for HCC survival prediction but also
highlight the broader potential of machine learning in enhancing prognostic accuracy beyond
traditional statistical methods. The study provides a framework for implementing these
algorithms in clinical decision support systems, with SVM particularly suited for settings
requiring high-precision classification, though RF remains advantageous when model
interpretability is prioritised. Future research should focus on integrating these machine
learning approaches with molecular data to develop more comprehensive prognostic tools.
Keywords: Hepatocellular carcinoma, Machine learning, Support Vector Machine, Random
Forest, XGBoost, Survival prediction, Clinical decision support.