Unsupervised Contextual Anomaly Detection

The goal of unsupervised contextual anomaly detection is to identify previously unseen rare objects or events in data that contains both behavioral attributes and contextual attributes, without any pre-stored information about anomalous observations. Behavioral attributes are directly related to the process of interest, while contextual attributes describe external but highly influential factors. This method achieves anomaly detection of behavioral attributes conditioned on contextual attributes through a joint deep variational generative model, which has significant application value.