Layered Learning for Early Anomaly Detection: Predicting Critical Health Episodes

Abstract

Critical health events represent a relevant cause of mortality in intensive care units of hospitals, and their timely prediction has been gaining increasing attention. This problem is an instance of the more general predictive task of early anomaly detection in time series data. One of the most common approaches to solve this problem is to use standard classification methods. In this paper we propose a novel method that uses a layered learning architecture to solve early anomaly detection problems. One key contribution of our work is the idea of pre-conditional events, which denote arbitrary but computable relaxed versions of the event of interest. We leverage this idea to break the original problem into two layers, which we hypothesize are easier to solve. Focusing on critical health episodes, the results suggest that the proposed approach is advantageous relative to state of the art approaches for early anomaly detection. Although we focus on a particular case study, the proposed method is generalizable to other domains.