Mining for Structural Anomalies in Graph-based Data

In this paper we present graph-based approaches to mining for anomalies in domains where the anomalies consist of unexpected entity/relationship alterations that closely resemble non-anomalous behavior. We introduce three novel algorithms for the purpose of detecting anomalies in all possible types of graph changes. Each of our algorithms focuses on a specific graph change and uses the minimum description length principle to discover those substructure instances that contain anomalous entities and relationships. Using synthetic and real-world data, we evaluate the effectiveness of each of these algorithms in terms of each of the types of anomalies. Each of these algorithms demonstrates the usefulness of examining a graph-based representation of data for the purposes of detecting fraud.