Imbalanced and semi-supervised classification for prognosis of ACLF

Abstract

Acute-on-chronic liver failure (ACLF) is characterized by jaundice, coagulopathy, hepatic encephalopathy, and associated with high mortality. According to the progress of patients, we partition 81 ACLF patients into three groups. Group I includes 40 improved patients, group II contains 18 death patients, and group III is composed of 23 unlabeled patients. For the imbalanced characteristic of groups I and II, we construct an imbalanced prediction model based on small sphere and large margin approach (SSLM). SSLM classifies two classes of samples by maximizing their margin and then is an effective classification method for imbalanced data. For groups I, II and III, we present a prediction model based on semi-supervised twin support vector machine (TSVM), which integrates 23 unlabeled samples into the training process and improves testing accuracy. Compared with other three algorithms, our two proposed prediction models produce better testing accuracy. Finally we apply them to predict 23 not confirmed patients, and integrate them with the MELD method to obtain their prediction labels.