This paper presents an application of machine learning in network management and supervision, in order to help the processing of the large volume of event notifications received by network operators. In this paper, we provide theoretical and experimental resutls on learning patterns called chronicles, in order to design a machine assistant to network supervision operators. We first defien a learning model that suits our framework, and study from a theoretical point of view the ability to learn chronicles. We quantify the effects of the network behaviour on learning and prove to what extent help can be brought by oracles, possibly the operator, or another learning assistant, to increase the assistnat accuracy. We also have implemented and tested our machine assistant and we give experimental resutls obtaiend in two distinct realworld situations. They show experimentally the circumstances for which chronicle learning is possible without the help of the operator or another assistant.
展开▼