eNodeB Failure Detection from Aggregated Performance KPIs in Smart-city LTE Infrastructures

In this paper we show how Supervised Binary Classification techniques can be used to tackle the problem of eNodeB failure detection in an LTE network carrying Machine-to-Machine (M2M) smart-city traffic. 22 different classifiers are trained with data from two 24 hrs simulations with different levels of traffic volume. Input features for the classification models are built aggregating packet generation and access collisions from the eNodeB on which failures are being detected, as well as from its closest neighbors, by computing statistics for each time-bin. Given that network service providers generally process real-time data to produce periodic aggregated summaries, we explore the effect of different levels of granularity in data aggregation and their effect on our ability to detect failures. The M2M traffic data was gathered from a simulated LTE network that uses publicly available geographic databases from the city of Montreal. With Linear Support Vector Machines (L-SVMs) and Bagged Decision Trees (BDT), failure detection rates above 97.5 % were achieved, with false positive rates under 2.8 %, showing that, even with 30 minutes aggregations, it is feasible to extract meaningful failure information.