Detecting M2M traffic in mobile cellular networks

Service visibility is a major part in traffic engineering and security. The recent rise of Machine-to-Machine Communication (M2M) nodes in cellular mobile networks and their impact on up-link resources draw the attention to automatic identification of this traffic class. However, the traditional traffic classification does not deliver accuracy the operators need. We present a method for detecting M2M traffic with an accuracy of up to 99% within the IP packet stream of a mobile operator. Traffic classification is based on features extracted from the packet level traces. Our main contribution is the extensive analysis of a large set of features where we showed that M2M can be classified very well using only nine features per node. In the supervised case we get a high level of accuracy starting at 2,5% of training data. In the unsupervised case we can cluster with a very good performance above 95% based on the extracted features. In this paper we are showing that it is possible to detecting M2M inside the traffic stream of a mobile cellular network at high accuracy, for both supervised and unsupervised machine learning.