SQoE KQIs anomaly detection in cellular networks: Fast online detection framework with Hourglass clustering

Abstract

The explosive growth of data volume in mobile networks makes fast online diagnose a pressing search problem. In this paper, an object-oriented detection framework with a two-step clustering, named as Hourglass Clustering, is given. Where three object parameters are chosen as Synthetical Quality of Experience (SQoE) Key Quality Indicators (KQIs) to reflect accessibility, integrality, and maintainability of networks. Then, we choose represented Key Performance Indicators (rKPIs) as cause parameters with correlation analysis. For these two kinds of parameters, a hybrid algorithm combining the self-organizing map (SOM) and k-medoids is used for clustering them into different types. We apply this framework to online anomaly detection in Cellular Networks, named SQoE-driven Anomaly Detection and Cause Location System (SQoE-ADCL). Our experiments with real 4G data show that besides fast online detection, SQoE-ADCL makes a better soft decision instead of a traditional hard decision. Furthermore, it is also a general way of being applied to other similar applications in big data.