MULTI-VIEW LEARNING BASED ON NON-REDUNDANT FUSION FOR ICU PATIENT MORTALITY PREDICTION

In medical data research, mortality prediction in intensive care units (ICUs) has always been a research hotspot. The Apache-II death prediction system relies on scoring rules. Despite its extensive application, it also has apparent shortcomings, with its accuracy rate decreasing over time. In recent years, researchers have proposed machine learning and deep learning algorithms to establish predictive models for ICUs. Those predicting from a single perspective cannot fully apply multiple sources of information, while the fusion of multiple perspectives may produce much redundant information. Therefore, this paper proposes a multi-view fusion method based on non-redundant information learning, applying it to ICU patient mortality prediction. Collaboratively, it applies consistency and complementarity among different views to discover internal data patterns accurately and improve the effectiveness of data analysis. Experimental results indicate that the accuracy of this method in predicting the mortality of critically ill patients in ICUs reaches 90.43%, around 5.25% higher than the existing model.