A robust algorithm for anomaly detection in mobile networks

Self-Organizing Network (SON) functions aim at automating mobile network management in three key areas: self-configuration, self-optimization and self-healing. This paper's focus is self-healing, the detection of outages and other faults in the network. The first step in every healing process is the detection of abnormal operation, also called anomaly or outlier detection. In this paper we propose a simple but effective statistics-based method for anomaly detection in mobile networks, using performance and failure Key Performance Indicators (KPI) to detect anomalies in cell behavior. The algorithm is aimed to be easy to setup, and is computationally less demanding than machine learning based algorithms, making it suitable for large networks or environments with low processing power. Results are presented on real mobile network data showing that the detector yields good results, outperforming basic failure indicator mechanisms.