Acute pancreatitis severity classifcation: accuracy, robustness, visualization

Abstract

The work is devoted to the problem of acute pancreatitis severity classifcation. This problem is characterized by a small amount of data, which leads to unstable estimations for new patients and a strong infuence of the training sample on the predictions. In this paper prediction stability visualization based on violin plot is proposed and applied. A simulation experiments are carried out to study the stability of linear regression, support vector machine, random forest trained with various subsets.