Automatic Detection of Stalling Events using Machine Learning Algorithms

Abstract

With the proliferation of new video streaming OTT service applications, content providers must guarantee an efficient, effective and satisfactory interaction between the user and the service application, to measure this they use the quality of experience (QoE). QoE in the context of telecommunications can be understood as the acceptability of a service perceived by end users. However, given the variability of the network and the different factors that can intervene during the service, this perception can be altered and one of these causes is stalling events. It is necessary to have an accurate method to detect stalling events, allowing content providers to make better decisions in the design of their OTT applications and thus reduce customer desertion and the economic losses of service or content providers. For this reason, in this paper firstly we show a comparative analysis of supervised algorithms showing its invalidity given the imbalance of the data despite the high levels of accuracy. Therfore, we propose as a contribution a novel method based on algorithms provided by SSAD (Semi Supervised Anomaly Detection) for the detection of stalling events. Finally, we observe that in our method the Isolation Forest model has the best performance, closer to 1 and we will show more detail about its performance in future works.