Discovering Anomalies to Multiple Normative Patterns in Structural and Numeric Data

One of the primary issues with traditional anomaly detection approaches is their inability to handle complex, structural data. One approach to this issue involves the detection of anomalies in data that is represented as a graph. The advantage of graph-based anomaly detection is that the relationships between elements can be analyzed, as opposed to just the data values themselves, for structural oddities in what could be a complex, rich set of information. However, until now, attempts at applying graph-based approaches to anomaly detection have encountered two issues: (1) Numeric values found in the data are not incorporated into the analysis of the structure, which could augment and improve the discovery of anomalies; and (2) The anomalous substructure may not be a deviation of the most prevalent pattern, but deviates from only one of many normative patterns. This paper presents enhancements to existing graph-based anomaly detection techniques that address these two issues and shows ...

• Type: Journal
• Year: 2009
• Where: FLAIRS
• Authors: William Eberle, Lawrence B. Holder
• Keywords: Anomaly Detection, Artificial Intelligence, FLAIRS 2009, Graph-based Anomaly Detection, Traditional Anomaly Detection