Comparative Evaluation of Machine Learning and Deep Learning Models for Early Prediction of Severe Acute Pancreatitis: A Multi-Model Study Using the 2012 Revised Atlanta Classification

The study compared 11 different machine learning and deep learning models for predicting severe acute pancreatitis (SAP) using routine laboratory values from patient admission. The research included 722 acute pancreatitis patients (585 with severe form, 137 with mild form) classified according to the 2012 Revised Atlanta Classification. The Random Forest model achieved the best results with an AUC of 0.877, sensitivity of 96.8%, and positive predictive value of 87.1%. Classical machine learning models significantly outperformed deep learning models, with the best deep learning model (CNN-LSTM) achieving only an AUC of 0.777. The results suggest that Random Forest can provide reliable early prediction of SAP severity using data available at patient admission. Further external prospective validation is required before clinical deployment.