Predicting Esophageal Varices in Cirrhotic Patients Using Machine Learning
The aim of this project was to assess the ability of machine learning models to predict esophageal varices in cirrhotic patients. A gradient boosting decision tree model; LightGBM was implemented to predict whether the cirrhotic patients have esophageal varices or not. Gender, presence of ascites, presence of encephalopathy, Child-Pugh Score and platelet counts are used as predictors, and they are selected after a feature selection process that consist of permutation feature importance and leave-one-out feature importance. Validation and uncertainty quantification was implemented with multiple shuffled train-test splits with replacement. This process was repeated 50 times for getting statistically significant results. Mean of test scores was the estimated error and standard deviation of test scores was the confidence interval. LightGBM model had 0.6853 (± 0.06 standard deviation) mean ROC AUC score and 0.7 F1 score. This project served as a baseline for the paper named “Artificial Intelligence to Predict Esophageal Varices in Patients with Cirrhosis”.