The identification of network applications through observation of associated packet traffic flows is vital to the areas of network management and surveillance. Currently popular methods such as port number and payload-based identification exhibit a number of shortfalls. An alternative is to use machine learning (ML) techniques and identify network applications based on per-flow statistics, derived from payload-independent features such as packet length and inter-arrival time distributions. The performance impact of feature set reduction, using Consistencybased and Correlation-based feature selection, is demonstrated on Naïve Bayes, C4.5, Bayesian Network and Naïve Bayes Tree algorithms. We then show that it is useful to differentiate algorithms based on computational performance rather than classification accuracy alone, as although classification accuracy between the algorithms is similar, computational performance can differ significantly.

Classification of network traffic using port-based or payload-based analysis is becoming increasingly difficult with many peer-to-peer (P2P) applications using dynamic port numbers, masquerading techniques, and encryption to avoid detection. An alternative approach is to classify traffic by exploiting the distinctive characteristics of applications when they communicate on a network. We pursue this latter approach and demonstrate how cluster analysis can be used to effectively identify groups of traffic that are similar using only transport layer statistics. Our work considers two unsupervised clustering algorithms, namely K-Means and DBSCAN, that have previously not been used for network traffic classification. We evaluate these two algorithms and compare them to the previously used AutoClass algorithm, using empirical Internet traces. The experimental results show that both K-Means and DBSCAN work very well and much more quickly then AutoClass. Our results indicate that although DBSCAN has lower accuracy compared to K-Means and AutoClass, DBSCAN produces better clusters.

The ubiquity of 802.11 devices and networks enables anyone to track our every move with alarming ease. Each 802.11 device transmits a globally unique and persistent MAC address and thus is trivially identifiable. In response, recent research has proposed replacing such identifiers with pseudonyms (i.e., temporary, unlinkable names). In this paper, we demonstrate that pseudonyms are insufficient to prevent tracking of 802.11 devices because implicit identifiers, or identifying characteristics of 802.11 traffic, can identify many users with high accuracy. For example, even without unique names and addresses, we estimate that an adversary can identify 64 % of users with 90 % accuracy when they spend a day at a busy hot spot. We present an automated procedure based on four previously unrecognized implicit identifiers that can identify users in three real 802.11 traces even when pseudonyms and encryption are employed. We find that the majority of users can be identified using our techniques, but our ability to identify users is not uniform; some users are not easily identifiable. Nonetheless, we show that even a single implicit identifier is sufficient to distinguish many users. Therefore, we argue that design considerations beyond eliminating explicit identifiers (i.e., unique names and addresses), must be addressed in order to prevent user tracking in wireless networks. Categories and Subject Descriptors:

Recent research on Internet traffic classification algorithms has yielded a flurry of proposed approaches for distinguishing types of traffic, but no systematic comparison of the various algorithms. This fragmented approach to traffic classification research leaves the operational community with no basis for consensus on what approach to use when, and how to interpret results. In this work we critically revisit traffic classification by conducting a thorough evaluation of three classification approaches, based on transport layer ports, host behavior, and flow features. A strength of our work is the broad range of data against which we test the three classification approaches: seven traces with payload collected in Japan, Korea, and the US. The diverse geographic locations, link characteristics and application traffic mix in these data allowed us to evaluate the approaches under a wide variety of conditions. We analyze the advantages and limitations of each approach, evaluate methods to overcome the limitations, and extract insights and recommendations for both the study and practical application of traffic classification. We make our software, classifiers, and data available for researchers interested in validating or extending this work. 1.

Identifying and categorizing network traffic by application type is challenging because of the continued evolution of applications, especially of those with a desire to be undetectable. The diminished effectiveness of port-based identification and the overheads of deep packet inspection approaches motivate us to classify traffic by exploiting distinctive flow characteristics of applications when they communicate on a network. In this paper, we explore this latter approach and propose a semi-supervised classification method that can accommodate both known and unknown applications. To the best of our knowledge, this is the first work to use semi-supervised learning techniques for the traffic classification problem. Our approach allows classifiers to be designed from training data that consists of only a few labeled and many unlabeled flows. We consider pragmatic classification issues such as longevity of classifiers and the need for retraining of classifiers. Our performance evaluation using empirical Internet traffic traces that span a 6-month period shows that: 1) high flow and byte classification accuracy (i.e., greater than 90%) can be achieved using training data that consists of a small number of labeled and a large number of unlabeled flows; 2) presence of “mice ” and “elephant ” flows in the Internet complicates the design of classifiers, especially of those with high byte accuracy, and necessities use of weighted sampling techniques to obtain training flows; and 3) retraining of classifiers is necessary only when there are non-transient changes in the network usage characteristics. As a proof of concept, we implement prototype offline and realtime classification systems to demonstrate the feasibility of our approach.

Literature on the use of machine learning (ML) algorithms for classifying IP traffic has relied on fullflows or the first few packets of flows. In contrast, many real-world scenarios require a classification decision well before a flow has finished even if the flow’s beginning is lost. This implies classification must be achieved using statistics derived from the most recent N packets taken at any arbitrary point in a flow’s lifetime. We propose training the classifier on a combination of short sub-flows (extracted from fullflow examples of the target application’s traffic). We demonstrate this optimisation using the Naïve Bayes ML algorithm, and show that our approach results in excellent performance even when classification is initiated mid-way through a flow with windows as small as 25 packets long. We suggest future use of unsupervised ML algorithms to identify optimal subflows for training. 1.

Abstract—Internet traffic identification is an important tool for network management. It allows operators to better predict future traffic matrices and demands, security personnel to detect anomalous behavior, and researchers to develop more realistic traffic models. We present here a traffic classifier that can achieve a high accuracy across a range of application types without any source or destination host-address or port information. We use supervised machine learning based on a Bayesian trained neural network. Though our technique uses training data with categories derived from packet content, training and testing were done using features derived from packet streams consisting of one or more packet headers. By providing classification without access to the contents of packets, our technique offers wider application than methods that require full packet/payloads for classification. This is a powerful advantage, using samples of classified traffic to permit the categorization of traffic based only upon commonly available information. Index Terms—Internet traffic, network operations, neural network applications, pattern recognition, traffic identification.

Flow records gathered by routers provide valuable coarse-granularity traffic information for several measurement-related network applications. However, due to high volumes of traffic, flow records need to be sampled before they are gathered. Current techniques for producing sampled flow records are either focused on selecting flows from which statistical estimates of traffic volume can be inferred, or have simplistic models for applications. Such sampled flow records are not suitable for many applications with more specific needs, such as ones that make decisions across flows. As a first step towards tailoring the sampling algorithm to an application’s needs, we design a generic language in which any particular application can express the classes of traffic of its interest. Our evaluation investigates the expressive power of our language, and whether flow records have sufficient information to enable sampling of records of relevance to applications. We use templates written in our custom language to instrument sampling tailored to three different applications—BLINC, Snort, and Bro. Our study, based on month-long datasets gathered at two different network locations, shows that by learning local traffic characteristics we can sample relevant flow records near-optimally with low false negatives in diverse applications.

Abstract—The identification of network applications that create traffic flows is vital to the areas of network management and surveillance. Current popular methods such as port number and payload-based identification are inadequate and exhibit a number of shortfalls. A potential solution is the use of machine learning techniques to identify network applications based on payload independent statistical features. In this paper we evaluate and compare the efficiency and performance of different feature selection and machine learning techniques based on flow data obtained from a number of public traffic traces. We also provide insights into which flow features are the most useful. Furthermore, we investigate the influence of other factors such as flow timeout and size of the training data set. We find significant performance differences between different algorithms and identify several algorithms that provide accurate (up to 99 % accuracy) and fast classification.

Abstract. We demonstrate that the browser implementation used at a host can be passively identified with significant precision and recall, using only coarse summaries of web traffic to and from that host. Our techniques utilize connection records containing only the source and destination addresses and ports, packet and byte counts, and the start and end times of each connection. We additionally provide two applications of browser identification. First, we show how to extend a network intrusion detection system to detect a broader range of malware. Second, we demonstrate the consequences of web browser identification to the deanonymization of web sites in flow records that have been anonymized.