The defense of PhD thesis of Maha MDINI will be held on September 20 th at 14:15 at IMT Atlantique, Rennes Campus, Petit Amphi. The thesis is entitled “Anomaly Detection and Root Cause Diagnosis in Cellular Networks”. It was made in the framework of a CIFRE thesis with Exfo (formely known as Astellia).
With the evolution of automation and artificial intelligence tools, mobile networks have become more and more machine reliant. Today, a large part of their management tasks runs in an autonomous way, without human intervention. The latest standards of the Third Generation Partnership Project (3GPP) aim at creating Self-Organizing Network (SON) where the processes of configuration, optimization and healing are fully automated. This work is about the healing process. This question have been studied by many researchers. They designed expert systems and applied Machine Learning (ML) algorithms in order to automate the healing process. However this question is still not fully addressed. A large part of the network troubleshooting still rely on human experts. For this reason, we have focused in this thesis on taking advantage of data analysis tools such as pattern recognition and statistical approaches to automate the troubleshooting task and carry it to a deeper level. The troubleshooting task is made up of three processes: detecting anomalies, analyzing their root causes and triggering adequate recovery actions. In this thesis, we focus on the two first objectives: anomaly detection and root cause diagnosis. The first objective is about detecting issues in the network automatically without including expert knowledge. To meet this objective, we have created an Anomaly Detection System (ADS) that learns autonomously from the network traffic and detects anomalies in real time in the flow of data. The algorithm we propose, Watchmen Anomaly Detection (WAD), is based on pattern recognition. The second objective is automatic diagnosis of network issues. This project aims at identifying the root cause of issues without any prior knowledge about the network topology and services. To address this question, we have designed an algorithm, Automatic Root Cause Diagnosis (ARCD) that identifies the roots of network issues. ARCD is composed of two independent threads: inefficiency Major Contributor identification and Incompatibility detection. WAD and ARCD have been proven to be effective. However, many improvements of these algorithms are possible. This thesis does not address fully the question of self-healing networks. Nevertheless, it contributes to the understanding and the implementation of this concept in production cellular networks.
