Innovative Predictive Modeling for Tailored Cirrhosis Management Unveiling Novel Feature

The severe scarring of the liver due to the scar tissue gradually replacing healthy liver tissue results in Cirrhosis. This leads to liver failure, portal hypertension, and liver cancer. Our research uses advanced machine learning models with innovative null handling and feature engineering strategies to enhance predictive accuracy. We apply techniques such as iterative SVM-based imputation and groupwise median imputation, leveraging group-specific statistics to handle missing data, alongside feature selection algorithms to extract informative features. Additionally, we propose the ALBI_SCORE, a novel feature based on clinical insights to boost predictive performance. Stratified 20-fold cross-validation was employed to fine-tune hyperparameters, ensuring robustness and generalizability. Key metrics such as Accuracy, Precision, Recall, and AUROC were used to evaluate the models. Our study found that the GroupWise median imputation technique achieved 90% accuracy, while the SVM-based imputation method achieved 84%. These results demonstrate the effectiveness of our approach in predicting cirrhosis outcomes.