NARGES: Prediction Model for Informed Routing in a Communications Network

There is a dependency between packet-loss and the delay and jitter time-series derived from a telecommunication link. Multimedia applications such as Voice over IP (VoIP) are sensitive to loss and packet recovery is not a merely efficient solution with the increasing number of Internet users. Predicting packet-loss from network dynamics of past transmissions is crucial to inform the next generation of routers in making smart decisions. This paper proposes a hybrid data mining model for routing management in a communications network, called NARGES. The proposed model is designed and implemented for predicting packetloss based on the forecasted delays and jitters. The model consists of two parts: a historical symbolic time-series approximation module, called HDAX, and a Multilayer Perceptron (MLP). It is validated with heterogeneous quality of service (QoS) datasets, namely delay, jitter and packet-loss time-series. The results show improved precision and quality of prediction compared to autoregressive moving average, ARMA.