Anomaly Detection in Streaming Time Series Data

Abstract

Anomalies can be the main carriers of significant and often critical information and the identification of these critical points can be the main purpose of many investigations in fields such as fraud detection, object tracking and environmental monitoring. Further, owing to rapid advances in data collection technology it has become increasingly common for organisations to be dealing with data that stream in large quantities. Therefore, the overall focus of this thesis is on detecting anomalies in streaming time series data. This thesis introduces three new algorithms for anomaly detection with special reference to their capabilities, competitive features and target applications.