Comparison of probabilistic models used for diagnosis in cellular networks

In the forthcoming years, different radio access technologies (GSM, GPRS, UMTS, etc.) will have to coexist within the same cellular network. In this scenario of increasingly complex networks, automated management is becoming a crucial issue to provide high-quality services. In this paper, a system for automatic fault diagnosis of the radio access part of a mobile communication system is presented. For this purpose, a probabilistic diagnosis model based on discrete Bayesian Networks (BNs) is proposed. There is always a trade-off between accuracy and complexity of the model. Hence, two alternative structures to code the dependencies among elements in the model are compared with regard to their simplicity and performance. Empirical results are examined, based on data from a live GSMIGPRS network. Taking into account the experiments, a BN structure is selected for diagnosis in cellular networks.