Anomalies detection in mobile network management data

Third generation (3G) mobile networks rely on distributed architectures where Operation and Maintenance Centers handle a large amount of information about network behavior. Such data can be processed to extract higher-level knowledge, useful for network management and optimization. In this paper we apply reduction techniques, such as Principal Component Analysis, to identify orthogonal subspaces representing the more interesting data contributing to overall variance and to split them up in "normal" and "anomalous" subspaces. Patterns within anomalous subspaces allow for early detection of network anomalies, improving mobile networks management and reducing the risk of malfunctioning.