TY - JOUR TI - Anomaly detection in network using non-negative matrix factorization techniques DO - https://doi.org/doi:10.7282/T33R0VV5 PY - 2015 AB - Anomaly detection is becoming an important problem in graph mining. This is because people are eager to find out unusual objects or patterns in a network, which may results in possible damages, emerging trends, or even creations in different types of graphs or networks. For example, the transaction occurs in out-of- hometown area with high amount may indicate credit card fraud in a bank transaction network; a substantial high frequency of connectivity in the network may infer possible web attack. The discoveries of these activities are important for people’s life and personal information security, and thus are important tasks for relevant institutions. There are mainly three kinds of anomalies in a graph data: node, edge and subgraph anomaly. This thesis presents new graph anomaly detection algorithms that provide scoring functions on detecting both node anomalies and edge anomalies based on non-negative matrix factorization techniques. The experiments on real life data verify that the suggested method could provide better performance in finding meaningful anomalies. KW - Industrial and Systems Engineering KW - Graph theory KW - Anomalies KW - Data mining LA - eng ER -