Anomaly Detection in Complex Spatiotemporal Networks Through Location Aware Geospatial Big Data Sets

Anomaly detection is an important problem that has been researched in several domains. Based on the available data patterns, various supervised and unsupervised anomaly detection techniques have been introduced. In this paper, a novel anomaly detection technique for location aware geospatial big dataset is outlined. Specifically, we focus on anomaly detection in spatiotemporal complex networks. The outlined technique incorporates components of anomaly quantification and decision making on spatiotemporal graphs and embeds simultaneous learning and detection procedures. The magnitude of an anomaly at each time step is quantified to signify the pattern of anomalous behavior in the spatiotemporal network. We illustrate the efficacy of the proposed method by detecting and indicating the time and location of a single or multiple anomalies in an illustrative traffic network problem. Theoretical experiments on a suite of six randomly generated traffic network problems have been performed. The performance of the proposed algorithm with tuned parameters on this random set of problem instances clearly establishes the effectiveness and applicability of the introduced solution procedure.