An Adaptive Fuzzy Clustering Technique for Traffic Prediction of Packet-switched Networks

Abstract

Traffic prediction is significant to QoS design because it assists efficient management of network resources to improve the reliability and performance of the next generation Internet. The unavoidable traffic variation caused by diverse Internet services complicates traffic prediction, particularly in a multi-hop network. To simplify the complicated statistical analysis used in traditional approaches, an adaptive traffic prediction approach featuring robustness, high accuracy and high adaptability is proposed in this paper. The proposed approach bases on a novel fuzzy clustering algorithm to generalize and unveil the hidden structure of traffic patterns. The unveiled structure represents the characteristics of the target traffic. Therefore, it can be referenced to predict traffic in a limited time period by fuzzy matching. To track the variation of target traffic, the proposed approach adopts an incremental and dynamic on-line clustering procedure so that the prediction can maintain high accuracy under traffic variation. To verify the performance of the proposed approach and investigate its properties, the periodical, Poisson and real video traffic patterns have been used to experiment. The experimental results showed an excellent performance of the developed adaptive predictor. The prediction errors, in average, are near 2.2%, 13.6% and 7.62% for periodical, Poisson and real video traffics, respectively.