Research Area:  Machine Learning
Network traffic classification is an important task for any current data network. There any many possible classification targets for the traffic, but we have considered as especially important the activity state of a connection and the identification of elephant flows (few connections carrying most of the traffic). With these detection targets, this work presents a modification of the gaNet architecture for classification. gaNet is an additive network model formed by learning blocks that are stacked iteratively following the principles of boosting models. The original gaNet model is intended for regression, being the purpose of this work to show that it can be extended to classification under several adaptations. The resulting architecture is a generic additive network applicable to any supervised classification problem (gaNet-C). To obtain experimental results, the model is applied to a type-of-traffic forecast problem using real IoT traffic from a mobile operator. The paper presents a comprehensive comparison of results between the proposed new model and many alternative algorithms in terms of classification and performance metrics. The proposed classifier can perform a k-step ahead detection forecast based exclusively on a limited time-series of previous values for each network connection. The results include two very different challenges: detection forecast of active connections and elephant flows; showing that, in both cases, the proposed algorithm presents state of the art results.
Author(s) Name:  Manuel Lopez-Martin,Belen Carro,Antonio Sanchez-Esguevillas
Journal name:  Future Generation Computer Systems
Publisher name:  Elsevier
Volume Information:  Volume 105, April 2020, Pages 331-345
Paper Link:   https://www.sciencedirect.com/science/article/abs/pii/S0167739X19322319