Logarithmic Loss in Machine Learning Models for Liver Cirrhosis Detection. (#1781)
Read ArticleDate of Conference
July 17-19, 2024
Published In
"Sustainable Engineering for a Diverse, Equitable, and Inclusive Future at the Service of Education, Research, and Industry for a Society 5.0."
Location of Conference
Costa Rica
Authors
Patiño-Pérez, Darwin
Molina-Calderón, Miguel
Ochoa-Flores, Ángel
Castro-Carrasco, José
Abstract
Accurately predicting the survival rate of patients with cirrhosis is very important in healthcare. In this study, we compared five classification models using Mayo Clinic data for primary biliary cirrhosis. The log loss score is used to evaluate the accuracy of the model in predicting survival. RandomForest shows the lowest log loss, followed by LogisticRegression and SVM with consistent prediction accuracy. On the other hand, Naive Bayes and kNN show more accurate results. K-fold cross-validation verifies the stability of the model. Limitations such as dataset dependency and lack of cirrhosis-specific studies were identified, indicating the need for future external validation and development of more accurate models applicable in clinical settings. In conclusion, RandomForest stands out for its high performance, but it is essential to carefully evaluate other models before clinical implementation to predict survival in cirrhotic patients.