LAUSR

Abstract Health status monitoring for critically ill patients can help medical stuff quickly discover and assess the changes of disease and then take appropriate treatment, which is of great clinical significance. In this study, a data-driven learning approach called LWPR-PCA is first applied to monitor health status of patients in intensive care unit (ICU). Locally weighted projection regression (LWPR) is used to approximate the complex nonlinear process by using local linear models, and then principal component analysis (PCA) is further applied to status monitoring. LWPR-PCA is a good candidate to establish an individual-type model for an ICU patient and to improve the global monitoring performance, which is the mainstream direction of modern medicine. To confirm the superiority of LWPR-PCA, physiological data of 18 ICU patients are collected, of which the mean fault detection rates (FDRs) are increased by 4.8% and 4.6%, and the mean fault alarm rates (FAR) are decreased by 6.7% and 5.9% in terms of two kinds of faults, compared to the latest reported method L-PCA, which combines just in time learning and modified PCA methods.