Synthetical QoE-Driven Anomalous Cell Pattern Detection with a Hybrid Algorithm

Abstract

Owing to more attention on quality of service (QoS) for subscribers, the mobile operators should shift evaluation standard from QoS to Quality of Experience (QoE). However, most researches in the field focus on end-to-end metrics, few of them consider synthetical QoE in the whole network. For mobile carriers, it is more significative to improve the overall system performance at the lowest cost. Therefore, the comprehensive evaluation of all users is more suitable for network optimization. As voice is still the basic service, we consider anomaly detection about voice service in this paper. Firstly, two synthetical QoE parameters, quality of voice (QoV) and successful rate of wireless access (WA), are considered to identify abnormalities of cells from the aspect of integrality and accessibility respectively. Then, we use a hybrid algorithm combining self-organizing map (SOM) and K-means to classify abnormal data points into several categories. After that, the data points for cells are treated as time series to compute the proportions in each anomaly model, which form anomalous cell patterns. To location where the exception happened accurately, the other 5 Key Performance Indicators (KPIs) are selected by association Rule according to the correlation between two synthetical QoE parameters. They are used to identify specific classes of faults. The experiment shows that the proposed method is effective to visualize and analyze anomalous cell patterns. It can be a guideline for the operators to perform faster and more efficient troubleshooting.