StreamSpot detects anomalous typed graphs arriving as a stream of individual edges by tracking an anomaly score for each graph.
StreamSpot is accurate, as exhibited on multiple syslog stream datasets with an average anomaly detection precision of 95%.
StreamSpot scales to over 10,000 edges/second, as witnessed on a commodity Xeon® machine running at 2.1GHz. It consumes bounded memory, as low as 250MB for a stream of ~25M edges.
Emaad Manzoor, Sadegh M. Milajerdi and Leman Akoglu. Fast Memory-efficient Anomaly Detection in Streaming Heterogeneous Graphs. In 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data mining (KDD). 2016.