Giovana Gelatti1, Pedro Pereira Rodrigues1,2
A large amount of data is continuously generated in intensive health care. An analysis of these streaming data can provide important information to improve the monitoring of the health conditions of patients. The volume, velocity and complexity of these data, which come unlabeled, make their analysis a challenging task. Machine learning techniques have been successfully used for data stream mining. The application of these techniques to intensive health care stream data can extract useful knowledge for health care monitoring, which includes the detection of changes in the behaviour of sensors, failures on machines or systems, and data anomalies, which might represent abnormal activities. Anomaly (or outlier) detection aims to nd exceptions or abnormalities in a dataset. Data exception, in medical context, can represent new disease pattern, an event to be further investigated, behaviour changes or possible health complications. Nonetheless, its analysis in data streams is a challenging task. Machine learning-based temporal abstractions deal with the management and abstraction of time based data, providing a high level of visualization of each data object in its context. Their use for anomaly detection may provide the medical expert a focus on relevant data and warnings. Other applications of machine learning techniques to data streams can support medical diagnosis, giving new, useful and relevant knowledge about the patients being monitored. The purpose of this paper is to review recent research in anomaly detection and temporal abstraction and how they have been applied to intensive care data streams.
keywords: Anomaly detection, outlier detection, temporal abstraction, intensive care