Caracterização e Previsão de Falhas em Serviços de Conectividade: uma Aplicação à Rede Ipê

Vitor F. Zanotelli, Giovanni Comarela, Rodolfo da Silva Villaça, Magnos Martinello

Abstract: The Ipê Network is essential for the Brazilian scientific community as it interconnects universities and research centers across the country. This paper analyzes some characteristics of the Ipê Network and explores the use of machine learning techniques for predicting failures in connectivity services using public data provided by the ViaIpê monitoring tool. The problem is modeled as a binary classification task using recurrent neural networks. The results show that the dependability of the connectivity service varies significantly in the different PoPs of the Ipê Network. In addition, despite the heterogeneity of this service, the prediction models are promising, presenting good accuracy and good precision in some scenarios.