Ring-Topology Echo State Networks for ICU Sepsis Classification

Abstract

Sepsis is a life threatening condition that can be treated if detected early. This paper presents a study of the application of a Ring Topology Echo State Network (ESN) algorithm to a sepsis prediction task based on ICU records. The implemented algorithm is compared with commonly used classifiers and a combination of both approaches. Finally, we address how different causal strategies on filling missing record values affected the final classification performances. Having a dataset with a limited number of time entries per patient, the utility score U = 0.188 obtained (team 51: PLUX) suggests that further research is needed in order for the ESN to capture the temporal dynamics of the problem at hand.