With the progress of technology and the increase in the demand of high speed mobile data, mobile networks keep growing both in size and complexity. Systems with such diversity and complexity can hardly be operated purely with human supervision, that is why automation of most of the operational tasks is an important area in getting prepared for the future of mobile networking. While most of the administrative tasks such as plug-and-play installation (self-configuration) and self-optimization are already implemented in the current mobile standards, self-healing, the third key area in having complete Self Organizing Networks, is still in early stages of development. The processing of the data that these systems provide, for maintenance or fault diagnosis is still mainly done by human supervisors. These operators are in possession of undocumented and hardly transferable expert knowledge, that makes automating these functions a difficult task.
This paper presents an automated diagnosis system that can identify different states in a mobile network through machine learning. The proposed system can diagnose errors or loss of performance using previously constructed expert knowledge data from the networks monitoring system, and can give good predictions of the cause of these problems. First, an overview is given of a modern mobile network architecture and a corresponding monitoring system that helps in understanding the original problem. The mobile network that this research is based on is the LTE (Long Term Evolution) third generation network. Since machine learning is not a widely known science, I try to give an overview of the possible methods that were taken into consideration during the design process of the diagnosis system. The paper describes the proposed system in detail, and shows the results of testing the system on data generated in a mobile network.