Content uploaded by Jeffrey Erman
Author content
All content in this area was uploaded by Jeffrey Erman on Oct 21, 2014
Content may be subject to copyright.
Traffic Classification Using Clustering Algorithms
Jeffrey Erman, Martin Arlitt, Anirban Mahanti
University of Calgary, 2500 University Drive NW, Calgary, AB, Canada
{erman, arlitt, mahanti}@cpsc.ucalgary.ca
ABSTRACT
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 tech-
niques, 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 eval-
uate these two algorithms and compare them to the previously used
AutoClass algorithm, using empirical Internet traces. The experi-
mental 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.
Categories and Subject Descriptors
I.5.4 [Computing Methodologies]: Pattern Recognition—Appli-
cations
General Terms
Algorithms, classification
Keywords
machine learning, unsupervised clustering
1. INTRODUCTION
Accurate identification and categorization of network traffic ac-
cording to application type is an important element of many net-
work management tasks such as flow prioritization, traffic shap-
ing/policing, and diagnostic monitoring. For example, a network
operator may want to identify and throttle (or block) traffic from
peer-to-peer (P2P) file sharing applications to manage its band-
width budget and to ensure good performance of business criti-
Permission to make digital or hard copies of all or part of this work for
personal or classroom use is granted without fee provided that copies are
not made or distributed for profit or commercial advantage and that copies
bear this notice and the full citation on the first page. To copy otherwise, to
republish, to post on servers or to redistribute to lists, requires prior specific
permission and/or a fee.
SIGCOMM’06 Workshops September 11-15, 2006, Pisa, Italy.
Copyright 2006 ACM 1-59593-417-0/06/0009 ...$5.00.
cal applications. Similar to network management tasks, many net-
work engineering problems such as workload characterization and
modelling, capacity planning, and route provisioning also benefit
from accurate identification of network traffic. In this paper, we
present preliminary results from our experience with using a ma-
chine learning approach called clustering for the network traffic
identification problem. In the remainder of this section, we moti-
vate why clustering is useful, discuss the specific contributions of
this paper, and outline our ongoing work.
The classical approach to traffic classification relies on mapping
applications to well-known port numbers and has been very suc-
cessful in the past. To avoid detection by this method, P2P appli-
cations began using dynamic port numbers, and also started dis-
guising themselves by using port numbers for commonly used pro-
tocols such as HTTP and FTP. Many recent studies confirm that
port-based identification of network traffic is ineffective [8, 15].
To address the aforementioned drawbacks of port-based classi-
fication, several payload-based analysis techniques have been pro-
posed [3, 6, 9, 11, 15]. In this approach, packet payloads are ana-
lyzed to determine whether they contain characteristic signatures of
known applications. Studies show that these approaches work very
well for the current Internet traffic including P2P traffic. In fact,
some commercial packet shaping tools have started using these
techniques. However, P2P applications such as BitTorrent are be-
ginning to elude this technique by using obfuscation methods such
as plain-text ciphers, variable-length padding, and/or encryption.
In addition, there are some other disadvantages. First, these tech-
niques only identify traffic for which signatures are available and
are unable to classify any other traffic. Second, these techniques
typically require increased processing and storage capacity.
The limitations of port-based and payload-based analysis have
motivated use of transport layer statistics for traffic classification [8,
10, 12, 14, 17]. These classification techniques rely on the fact
that different applications typically have distinct behaviour patterns
when communicating on a network. For instance, a large file trans-
fer using FTP would have a longer connection duration and larger
average packet size than an instant messaging client sending short
occasional messages to other clients. Similarly, some P2P appli-
cations such as BitTorrent 1can be distinguished from FTP data
transfers because these P2P connections typically are persistent
and send data bidirectionally; FTP data transfer connections are
non-persistent and send data only unidirectionally. Transport layer
statistics such as the total number of packets sent, the ratio of the
bytes sent in each direction, the duration of the connection, and the
average size of the packets characterize these behaviours.
In this paper, we explore the use of a machine learning approach
called clustering for classifying traffic using only transport layer
1http://www.bittorrent.org/protocol.html
statistics. Cluster analysis is one of the most prominent methods
for identifying classes amongst a group of objects, and has been
used as a tool in many fields such as biology, finance, and com-
puter science. Recent work by McGregor et al. [10] and Zander
et al. [17] show that cluster analysis has the ability to group Inter-
net traffic using only transport layer characteristics. In this paper,
we confirm their observations by evaluating two clustering algo-
rithms, namely K-Means [7] and DBSCAN [5], that to the best of
our knowledge have not been previously applied to this problem.
In addition, as a baseline, we present results from the previously
considered AutoClass [1] algorithm [10, 17].
The algorithms evaluated in this paper use an unsupervised learn-
ing mechanism, wherein unlabelled training data is grouped based
on similarity. This ability to group unlabelled training data is ad-
vantageous and offers some practical benefits over learning ap-
proaches that require labelled training data (discussed in Section
2). Although the selected algorithms use an unsupervised learning
mechanism, each of these algorithms, however, is based on differ-
ent clustering principles. The K-Means clustering algorithm is a
partition-based algorithm [7], the DBSCAN algorithm is a density-
based algorithm [5], and the AutoClass algorithm is a probabilistic
model-based algorithm [1]. One reason in particular why K-Means
and DBSCAN algorithms were chosen is that they are much faster
at clustering data than the previously used AutoClass algorithm.
We evaluate the algorithms using two empirical traces: a well-
known publicly available Internet traffic trace from the University
of Auckland, and a recent trace we collected from the University
of Calgary’s Internet connection. The algorithms are compared
based on their ability to generate clusters that have a high predictive
power of a single application. We show that clustering works for
a variety of different applications, including Web, P2P file-sharing,
and file transfer with the AutoClass and K-Means algorithm’s ac-
curacy exceeding 85% in our results and DBSCAN achieving an
accuracy of 75%. Furthermore, we analyze the number of clusters
and the number of objects in each of the clusters produced by the
different algorithms. In general, the ability of an algorithm to group
objects into a few “good” clusters is particularly useful in reducing
the amount of processing required to label the clusters. We show
that while DBSCAN has a lower overall accuracy the clusters it
forms are the most accurate. Additionally, we find that by looking
at only a few of DBSCAN’s clusters one could identify a significant
portion of the connections.
Ours is a work-in-progress. Preliminary results indicate that
clustering is indeed a useful technique for traffic identification. Our
goal is to build an efficient and accurate classification tool using
clustering techniques as the building block. Such a clustering tool
would consist of two stages: a model building stage and a classifi-
cation stage. In the first stage, an unsupervised clustering algorithm
clusters training data. This produces a set of clusters that are then
labelled to become our classification model. In the second stage,
this model is used to develop a classifier that has the ability to label
both online and offline network traffic. We note that offline classifi-
cation is relatively easier compared to online classification, as flow
statistics needed by the clustering algorithm may be easily obtained
in the former case; the latter requires use of estimation techniques
for flow statistics. We should also note that this approach is not a
“panacea” for the traffic classification problem. While the model
building phase does automatically generate clusters, we still need
to use other techniques to label the clusters (e.g., payload anal-
ysis, manual classification, port-based analysis, or a combination
thereof). This task is manageable because the model would typi-
cally be built using small data sets.
We believe that in order to build an accurate classifier, a good
classification model must be used. In this paper, we focused on the
model building step. Specifically, we investigate which clustering
algorithm generates the best model. We are currently investigating
building efficient classifiers for K-Means and DBSCAN and testing
the classification accuracy of the algorithms. We are also investi-
gating how often the models should be retrained (e.g., on a daily,
weekly, or monthly basis).
The remainder of this paper is arranged as follows. The different
Internet traffic classification methods including those using cluster
analysis are reviewed in Section 2. Section 3 outlines the theory
and methods employed by the clustering algorithms studied in this
paper. Section 4 and Section 5 present our methodology and out-
line our experimental results, respectively. Section 6 discusses the
experimental results. Section 7 presents our conclusions.
2. BACKGROUND
Several techniques use transport layer information to address the
problems associated with payload-based analysis and the diminish-
ing effectiveness of port-based identification. McGregor et al. hy-
pothesize the ability of using cluster analysis to group flows using
transport layer attributes [10]. The authors, however, do not evalu-
ate the accuracy of the classification as well as which flow attributes
produce the best results. Zander et al. extend this work by using
another Expectation Maximization (EM) algorithm [2] called Au-
toClass [1] and analyze the best set of attributes to use [17]. Both
[10] and [17] only test Bayesian clustering techniques implemented
by an EM algorithm. The EM algorithm has a slow learning time.
This paper evaluates clustering algorithms that are different and
faster than the EM algorithm used in previous work.
Some non-clustering techniques also use transport layer statis-
tics to classify traffic [8, 9, 12, 14]. Roughan et al. use nearest
neighbor and linear discriminate analysis [14]. The connection du-
rations and average packet size are used for classifying traffic into
four distinct classes. This approach has some limitations in that the
analysis from these two statistics may not be enough to classify all
applications classes.
Karagiannis et al. propose a technique that uses the unique be-
haviors of P2P applications when they are transferring data or mak-
ing connections to identify this traffic [8]. Their results show that
this approach is comparable with that of payload-based identifica-
tion in terms of accuracy. More recently, Karagiannis et al. devel-
oped another method that uses the social, functional, and applica-
tion behaviors to identify all types of traffic [9]. These approaches
focus on higher level behaviours such as the number of concurrent
connections to an IP address and does not use the transport layer
characteristics of single connection that we utilize in this paper.
In [12], Moore et al. use a supervised machine learning algo-
rithm called Na¨ıve Bayes as a classifier. Moore et al. show that the
Na¨ıve Bayes approach has a high accuracy classifying traffic. Su-
pervised learning requires the training data to be labelled before the
model is built. We believe that an unsupervised clustering approach
offers some advantages over supervised learning approaches. One
of the main benefits is that new applications can be identified by
examining the connections that are grouped to form a new clus-
ter. The supervised approach can not discover new applications
and can only classify traffic for which it has labelled training data.
Another advantage occurs when the connections are being labelled.
Due to the high accuracy of our clusters, only a few of the connec-
tions need to be identified in order to label the cluster with a high
degree of confidence. Also consider the case where the data set be-
ing clustered contains encrypted P2P connections or other types of
encrypted traffic. These connections would not be labelled using
payload-based classification. These connections would, therefore,
be excluded from the supervised learning approach which can only
use labelled training data as input. This could reduce the super-
vised approach’s accuracy. However, the unsupervised clustering
approach does not have this limitation. It might place the encrypted
P2P traffic into a cluster with other unencrypted P2P traffic. By
looking at the connections in the cluster, an analyst may be able to
see similarities between unencrypted P2P traffic and the encrypted
traffic and conclude that it may be P2P traffic.
3. CLUSTERING ALGORITHMS
This section reviews the clustering algorithms, namely K-Means,
DBSCAN, and AutoClass, considered in this work. The K-Means
algorithm produces clusters that are spherical in shape whereas the
DBSCAN algorithm has the ability to produce clusters that are non-
spherical. The different cluster shapes that DBSCAN is capable
of finding may allow for a better set of clusters to be found that
minimize the amount of analysis required. The AutoClass algo-
rithm uses a Bayesian approach and can automatically determine
the number of clusters. Additionally, it performs soft clustering
wherein objects are assigned to multiple clusters fractionally.
The Cluster 3.0 [4] software suite is used to obtain the results
for K-Means clustering. The DBSCAN results are obtained the
WEKA software suite [16]. The AutoClass results are obtained
using an implementation provided by [1].
In order for the clustering of the connections to occur, a similar-
ity (or distance) measurement must be established first. While vari-
ous similarity measurements exist, Euclidean distance is one of the
most commonly used metrics for clustering problems [7, 16]. With
Euclidean distance, a small distance between two objects implies a
strong similarity whereas a large distance implies a low similarity.
In an n-dimensional space of features, Euclidean distance can be
calculated between objects xand yas follows:
dist (x, y) = v
u
u
t
n
X
i=1
(xi−yi)2,(1)
with nbeing the number of features in each object. The algorithms
in this paper all use the Euclidean distance as their similarity mea-
surement. The objects in this case will be connections and the fea-
tures are the connection’s transport layer statistics.
3.1 K-Means Clustering
There are a variety of partition-based clustering algorithms avail-
able [7]. The K-Means algorithm is selected because it is one of the
quickest and most simple. The K-Means algorithm partitions ob-
jects in a data set into a fixed number of K disjoint subsets. For
each cluster, the partitioning algorithm maximizes the homogene-
ity within the cluster by minimizing the square-error. The formula
for the square error is:
E=
K
X
i=1
n
X
j=1
|dist(xj, ci)|2.(2)
The square error is calculated as the distance squared between each
object xand the centre (or mean) of its cluster. Object crepresents
the respective centre of each cluster.
The square error is minimized by K-Means using the following
algorithm. The centers of the K clusters are initially chosen ran-
domly from within the subspace. The objects in the data set are then
partitioned into the nearest cluster. K-Means iteratively computes
the new centers of the clusters that are formed and then repartitions
them based on the new centers. The K-Means algorithm continues
this process until the membership within the clusters stabilizes, thus
producing the final partitioning. The algorithm converges within a
small number of iterations for the data sets tested in this paper.
3.2 DBSCAN Clustering
Density-based algorithms regard clusters as dense areas of ob-
jects that are separated by less dense areas. These clustering algo-
rithms have an advantage over partition-based algorithms because
they are not limited to finding spherical shaped clusters but can
find clusters of arbitrary shapes. In this paper, the DBSCAN (Den-
sity Based Spatial Clustering of Applications with Noise) algorithm
was chosen as a representative of density-based algorithms [5].
The DBSCAN algorithm is based on the concepts of density-
reachability and density-connectivity. These concepts depend on
two input parameters: epsilon (eps) and minimum number of points
(minPts). Epsilon is the distance around an object that defines its
eps-neighborhood. For a given object q, when the number of ob-
jects within the eps-neighborhood is at least minPts, then qis de-
fined as a core object. All objects within its eps-neighborhood are
said to be directly density-reachable from q. In addition, an object
pis said to be density-reachable if it is within the eps-neighborhood
of an object that is directly density-reachable or density-reachable
from q. Furthermore, objects pand qare said to be density-connected
if an object oexists that both pand qare density-reachable from.
These notions of density-reachability and density-connectivity
are used to define what the DBSCAN algorithm considers as a clus-
ter. A cluster is defined as the set of objects in a data set that are
density-connected to a particular core object. Any object that is
not part of a cluster is categorized as noise. This is in contrast to
K-Means and AutoClass, which assign every object to a cluster.
The DBSCAN algorithm works as follows. Initially, all ob-
jects in the data set are assumed to be unassigned. DBSCAN then
chooses an arbitrary unassigned object pfrom the data set. If DB-
SCAN finds pis a core object, it finds all the density-connected
objects based on eps and minPts. It assigns all these objects to a
new cluster. If DBSCAN finds pis not a core object, then pis con-
sidered to be noise and DBSCAN moves onto the next unassigned
object. Once every object is assigned, the algorithm stops.
3.3 AutoClass
Probabilistic model-based clustering, previously considered in
[10, 17], is another powerful clustering technique. We use an im-
plementation of a probabilistic model-based clustering technique
called AutoClass [1]. This algorithm allows for the automatic se-
lection of the number of clusters and the soft clustering of the data.
Soft clusters allow the data objects to be fractionally assigned to
more than one cluster. For our work, we use the most probable
assignment as the object’s assignment.
To build the probabilistic model, the clustering algorithm deter-
mines the number of clusters and the parameters that govern the dis-
tinct probability distributions of each cluster. To accomplish this,
AutoClass uses the Expectation Maximization (EM) algorithm [2].
The EM algorithm has two steps: an expectation step and a max-
imization step. The initial expectation step guesses what the pa-
rameters are using pseudo-random numbers. In the maximization
step, the mean and variance are used to re-estimate the parameters
continually until they converge to a local maximum. These local
maxima are recorded and the EM process is repeated. This process
continues until enough samples of the parameters have been found
(we use 200 cycles in our results). AutoClass uses a Bayesian score
to determine the best set of parameters to use for the probabilistic
model. The Bayesian score is based on intra-cluster similarity and
inter-cluster dissimilarity. Also, the Bayesian score penalizes mod-
els with more clusters to minimize potential over-fitting.
4. METHODOLOGY
4.1 Empirical Traces
To analyze the algorithms, we used data from two empirical
packet traces. One is a publicly available packet trace called Auck-
land IV2, the other is a full packet trace that we collected ourselves
at the University of Calgary.
Auckland IV: The Auckland IV trace contains only TCP/IP head-
ers of the traffic going through the University of Auckland’s link
to the Internet. We used a subset of the Auckland IV trace from
March 16, 2001 at 06:00:00 to March 19, 2001 at 05:59:59. This
subset provided sufficient connection samples to build our model
(see Section 4.4).
Calgary: This trace was collected from a traffic monitor attached
to the University of Calgary’s Internet link. We collected this trace
on March 10, 2006 from 1 to 2pm. This trace is a full packet trace
with the entire payloads of all the packets captured. Due to the
amount of data generated when capturing full payloads, the disk
capacity (60 GB) of our traffic monitor was filled after one hour of
collection, thus, limiting the duration of the trace.
4.2 Connection Identification
To collect the statistical flow information necessary for the clus-
tering evaluations, the flows must be identified within the traces.
These flows, also known as connections, are a bidirectional ex-
change of packets between two nodes.
In the traces, the data is not exclusively from connection-based
transport layer protocols such as TCP. While this study focused
solely on the TCP-based applications it should be noted that statis-
tical flow information could be calculated for UDP traffic also. We
identified the start of a connection using TCP’s 3-way handshake
and terminated a connection when FIN/RST packets were received.
In addition, we assumed that a flow is terminated if the connection
was idle for over 90 seconds.
The statistical flow characteristics considered include: total num-
ber of packets, mean packet size, mean payload size excluding
headers, number of bytes transfered (in each direction and com-
bined), and mean inter-arrival time of packets. Our decision to use
these characteristics was based primarily on the previous work done
by Zander et al. [17]. Due the heavy-tail distribution of many of the
characteristics and our use of Euclidean distance as our similarity
metric, we found that the logarithms of the characteristics gives
much better results for all the clustering algorithms [13, 16].
4.3 Classification of the Data Sets
The publicly available Auckland IV traces include no payload
information. Thus, to determine the connections “true” classifica-
tions port numbers are used. For this trace, we believe that a port-
based classification will be largely accurate, as this archived trace
predates the widespread use of dynamic port numbers. The classes
considered for the Auckland IV datasets are DNS, FTP (control),
FTP (data), HTTP, IRC, LIMEWIRE, NNTP, POP3, and SOCKS.
LimeWire is a P2P application that uses the Gnutella protocol.
In the Calgary trace, we were able to capture the full payloads of
the packets, and therefore, were able to use an automated payload-
based classification to determine the “true” classes. The payload-
based classification algorithm and signatures we used is very sim-
ilar to those described by Karagiannis et al. [9]. We augmented
their signatures to classify some newer P2P applications and instant
messaging programs. The traffic classes considered for the Calgary
trace are HTTP, P2P, SMTP, and POP3. The application breakdown
2Available at: http://www.wand.net.nz/wand/wits/auck/
Table 1: Application breakdown of the Calgary trace
Application Connections Bytes % Bytes
HTTP 1,132,920 23,693,723,103 47.3%
P2P 41,478 17,578,995,934 35.1%
SMTP 46,882 2,997,244,939 6.0%
IMAP 2,955 228,156,060 0.5%
POP3 3,674 72,274,560 0.1%
MSSQL 8,105 23,824,936 0.0%
OTHER 41,239 658,046,156 1.3%
UNKNOWN 354,798 4,811,332,761 9.6%
of the Calgary trace is presented in Table 1. The breakdown of the
Auckland IV trace has been omitted due to space limitations. How-
ever, HTTP is also the most dominant application accounting for
over 76% of the bytes and connections.
4.4 Testing Methodology
The majority of the connections in both traces carry HTTP traf-
fic. This unequal distribution does not allow for equal testing of the
different classes. To address this problem, the Auckland data sets
used for the clustering consist of 1000 random samples of each traf-
fic class, and the Calgary data sets use 2000 random sample of each
traffic category. This allows the test results to fairly judge the abil-
ity on all traffic and not just HTTP. The size of the data sets were
limited to 8000 connections because this was the upper bound that
the AutoClass algorithm could cluster within a reasonable amount
of time (4-10 hours). In addition, to achieve a greater confidence in
the results we generated 10 different data sets for each trace. Each
of these data sets was then, in turn, used to evaluate the cluster-
ing algorithms. We report the minimum, maximum, and average
results from the data sets of each trace.
In the future, we plan on examining the practical issue of what
is the best way to pick the connections used as samples to build
the model. Some ways that we think this could be accomplished is
by random selection or a weighted selection using different criteria
such as bytes transfered or duration. Also, in order to get a reason-
able representative model of the traffic, one would need to select
a fairly large yet manageable number of samples. We found that
K-Means and DBSCAN algorithms are able to cluster much larger
data sets (greater than 100,000) within 4-10 hours.
5. EXPERIMENTAL RESULTS
In this section, the overall effectiveness of each clustering algo-
rithm is evaluated first. Next, the number of objects in each cluster
produced by the algorithms are analyzed.
5.1 Algorithm Effectiveness
The overall effectiveness of the clustering algorithms is calcu-
lated using overall accuracy. This overall accuracy measurement
determines how well the clustering algorithm is able to create clus-
ters that contain only a single traffic category.
The traffic class that makes up the majority of the connections
in a cluster is used to label the cluster. The number of correctly
classified connections in a cluster is referred to as the True Pos-
itives (TP). Any connections that are not correctly classified are
considered False Positives (FP). Any connection that has not been
assigned to a cluster is labelled as noise. The overall accuracy is
thus calculated as follows:
overall accuracy =PT P for all clusters
total number of connections .(3)
In the following subsections, the effectiveness of the K-Means,
DBSCAN, and AutoClass algorithms are presented.
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 20 40 60 80 100 120 140 160
Overall Accuracy
Number of Clusters
Calgary
AucklandIV
Figure 1: Accuracy using K-Means
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 0.01 0.02 0.03 0.04
Overall Accuracy
Epsilon Distance
Auckland IV (3 minPts)
Calgary (3 minPts)
Figure 2: Accuracy using DBSCAN
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 0.01 0.02 0.03 0.04
Overall Accuracy
Epsilon Distance
3 minPts
6 minPts
12 minPts
24 minPts
Figure 3: Parametrization of DBSCAN
Table 2: Accuracy using AutoClass
Data Set Average Minimum Maximum
Auckland IV 92.4% 91.5% 93.5%
Calgary 88.7% 86.6% 90.0%
5.1.1 K-Means Clustering
The K-Means algorithm has an input parameter of K. This input
parameter as mentioned in Section 3.1, is the number of disjoint
partitions used by K-Means. In our data sets, we would expect
there would be at least one cluster for each traffic class. In ad-
dition, due to the diversity of the traffic in some classes such as
HTTP (e.g., browsing, bulk download, streaming) we would ex-
pect even more clusters to be formed. Therefore, based on this, the
K-Means algorithm was evaluated with K initially being 10 and K
being incremented by 10 for each subsequent clustering. The min-
imum, maximum, and average results for the K-Means clustering
algorithm are shown in Figure 1.
Initially, when the number of clusters is small the overall ac-
curacy of K-Means is approximately 49% for the Auckland IV
data sets and 67% for the Calgary data sets. The overall accuracy
steadily improves as the number of clusters increases. This contin-
ues until K is around 100 with the overall accuracy being 79% and
84% on average, for the Auckland IV and Calgary data sets, respec-
tively. At this point, the improvement is much more gradual with
the overall accuracy only improving by an additional 1.0% when K
is 150 in both data sets. When K is greater than 150, the improve-
ment is further diminished with the overall accuracy improving to
the high 80% range when K is 500. However, large values of K
increase the likelihood of over-fitting.
5.1.2 DBSCAN Clustering
The accuracy results for the DBSCAN algorithm are presented in
Figure 2. Recall that DBSCAN has two input parameters (minPts,
eps). We varied these parameters, and in Figure 2 report results
for the combination that produce the best clustering results. The
values used for minPts were tested between 3 and 24. The eps dis-
tance was tested from 0.005 to 0.040. Figure 3 presents results for
different combinations of (minPts, eps) values for the Calgary data
sets. As may be expected, when the minPts was 3 better results
were produced than when the minPts was 24 because smaller clus-
ters are formed. The additional clusters found using three minPts
were typically small clusters containing only 3 to 5 connections.
When using minPts equal to 3 while varying the eps distance
between 0.005 and 0.020 (see Figure 2), the DBSCAN algorithm
improved its overall accuracy from 59.5% to 75.6% for the Auck-
land IV data sets. For the Calgary data sets, the DBSCAN algo-
rithm improved its overall accuracy from 32.0% to 72.0% as the
eps distance was varied with these same values. The overall ac-
curacy for eps distances greater than 0.020 decreased significantly
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
% Connections
% Clusters
DBSCAN
K-Means
AutoClass
Figure 4: CDF of cluster weights
as the distance increased. Our analysis indicates that this large de-
crease occurs because the clusters of different traffic classes merge
into a single large cluster. We found that this larger cluster was for
connections with few packets, few bytes transfered, and short dura-
tions. This cluster contained typically equal amounts of P2P, POP3,
and SMTP connections. Many of the SMTP connections were for
emails with rejected recipient addresses and connections immedi-
ately closed after connecting to the SMTP server. For POP3, many
of the connections contained instances where no email was in the
users mailbox. Gnutella clients attempting to connect to a remote
node and having its “GNUTELLA CONNECT” packets rejected
accounted for most of the P2P connections.
5.1.3 AutoClass Clustering
The results for the AutoClass algorithm are shown in Table 2.
For this algorithm, the number of clusters and the cluster param-
eters are automatically determined. Overall, the AutoClass algo-
rithm has the highest accuracy. On average, AutoClass is 92.4%
and 88.7% accurate in the Auckland IV and Calgary data sets, re-
spectively. AutoClass produces an average of 167 clusters for the
Auckland IV data sets, and 247 clusters for the Calgary data sets.
5.2 Cluster Weights
For the traffic classification problem, the number of clusters pro-
duced by a clustering algorithm is an important consideration. The
reason being that once the clustering is complete, each of the clus-
ters must be labelled. Minimizing the number of clusters is also
cost effective during the classification stage.
One way of reducing the number of clusters to label is by evalu-
ating the clusters with many connections in them. For example, if a
clustering algorithm with high accuracy places the majority of the
connections in a small subset of the clusters, then by analyzing only
this subset a majority of the connections can be classified. Figure 4
shows the percentage of connections represented as the percentage
of clusters increases, using the Auckland IV data sets. In this eval-
uation, the K-Means algorithm had 100 for K. For the DBSCAN
and AutoClass algorithms, the number of clusters can not be set.
0.5
0.6
0.7
0.8
0.9
1
HTTP P2P POP3 SMTP
Precision
K-Means
DBSCAN
AutoClass
Figure 5: Precision using DBSCAN, K-Means, and AutoClass
DBSCAN uses 0.03 for eps, 3 for minPts, and has, on average,
190 clusters. We selected this point because it gave the best overall
accuracy for DBSCAN. AutoClass has, on average, 167 clusters.
As seen in Figure 4, both K-Means and AutoClass have more
evenly distributed clusters than DBSCAN. The 15 largest clusters
produced by K-Means only contain 50% of the connections. In
contrast, for the DBSCAN algorithm the five largest clusters con-
tain over 50% of the connections in the data sets. These five clus-
ters identified 75.4% of the NNTP, POP3, SOCKS, DNS, and IRC
connections with a 97.6% overall accuracy. These results are un-
expected when considering that by only looking at five of the 190
clusters, one can identify a significant portion of traffic. Qualita-
tively similar results were obtained for the Calgary data sets.
6. DISCUSSION
The DBSCAN algorithm is the only algorithm considered in this
paper that can label connections as noise. The K-Means and Au-
toClass algorithms place every connection into a cluster. The con-
nections that are labelled as noise reduce the overall accuracy of
the DBSCAN algorithm because they are regarded as misclassified.
We have found some interesting results by excluding the connec-
tions labelled as noise and just examining the clusters produced by
DBSCAN. Figure 5 shows the precision values for the DBSCAN
(eps=0.02, minPts=3), the K-Means (K=190), and the AutoClass
algorithms using the Calgary data sets. Precision is the ratio of TP
to FP for a traffic class. Precision measures the accuracy of the
clusters to classify a particular category of traffic.
Figure 5 shows that for the Calgary data sets, the DBSCAN algo-
rithm has the highest precision values for three of the four classes
of traffic. While not shown for the Auckland IV data sets, seven
of the nine traffic classes have average precision values over 95%.
This shows that while DBSCAN’s overall accuracy is lower than
K-Means and AutoClass it produces highly accurate clusters.
Another noteworthy difference among the clustering algorithms
is the time required to build the models. On average to build the
models, the K-Means algorithm took 1 minute, the DBSCAN algo-
rithm took 3 minutes, and the AutoClass algorithm took 4.5 hours.
Clearly, the model building phase of AutoClass is time consum-
ing. We believe this may deter systems developers from using this
algorithm even if the frequency of retraining the model is low.
7. CONCLUSIONS
In this paper, we evaluated three different clustering algorithms,
namely K-Means, DBSCAN, and AutoClass, for the network traffic
classification problem. Our analysis is based on each algorithm’s
ability to produce clusters that have a high predictive power of a
single traffic class, and each algorithm’s ability to generate a min-
imal number of clusters that contain the majority of the connec-
tions. The results showed that the AutoClass algorithm produces
the best overall accuracy. However, the DBSCAN algorithm has
great potential because it places the majority of the connections in
a small subset of the clusters. This is very useful because these
clusters have a high predictive power of a single category of traffic.
The overall accuracy of the K-Means algorithm is only marginally
lower than that of the AutoClass algorithm, but is more suitable for
this problem due to its much faster model building time. Ours in
a work-in-progress and we continue to investigate these and other
clustering algorithms for use as an efficient classification tool.
8. ACKNOWLEDGMENTS
This work was supported by the Natural Sciences and Engineer-
ing Research Council (NSERC) of Canada and Informatics Circle
of Research Excellence (iCORE) of the province of Alberta.
We thank Carey Williamson for his comments and suggestions
which helped improve this paper.
9. REFERENCES
[1] P. Cheeseman and J. Strutz. Bayesian Classification (AutoClass):
Theory and Results. In Advances in Knowledge Discovery and Data
Mining, AAI/MIT Press, USA, 1996.
[2] A. P. Dempster, N. M. Paird, and D. B. Rubin. Maximum likelihood
from incomeplete data via the EM algorithm. Journal of the Royal
Statistical Society, 39(1):1–38, 1977.
[3] C. Dews, A. Wichmann, and A. Feldmann. An analysis of internet
chat systems. In IMC’03, Miami Beach, USA, Oct 27-29, 2003.
[4] M. B. Eisen, P. T. Spellman, P. O. Brown, and D. Botstein. Cluster
Analysis and Display of Genome-wide Expression Patterns.
Genetics, 95(1):14863–15868, 1998.
[5] M. Ester, H. Kriegel, J. Sander, and X. Xu. A Density-based
Algorithm for Discovering Clusters in Large Spatial Databases with
Noise. In 2nd Int. Conf. on Knowledge Discovery and Data Mining
(KDD 96), Portland, USA, 1996.
[6] P. Haffner, S. Sen, O. Spatscheck, and D. Wang. ACAS: Automated
Construction of Application Signatures. In SIGCOMM’05 MineNet
Workshop, Philadelphia, USA, August 22-26, 2005.
[7] A. K. Jain and R. C. Dubes. Algorithms for Clustering Data. Prentice
Hall, Englewood Cliffs, USA, 1988.
[8] T. Karagiannis, A. Broido, M. Faloutsos, and K. claffy. Transport
Layer Identification of P2P Traffic. In IMC’04, Taormina, Italy,
October 25-27, 2004.
[9] T. Karagiannis, K. Papagiannaki, and M. Faloutsos. BLINK:
Multilevel Traffic Classification in the Dark. In SIGCOMM’05,
Philadelphia, USA, August 21-26, 2005.
[10] A. McGregor, M. Hall, P. Lorier, and J. Brunskill. Flow Clustering
Using Machine Learning Techniques. In PAM 2004, Antibes
Juan-les-Pins, France, April 19-20, 2004.
[11] A. W. Moore and K. Papagiannaki. Toward the Accurate
Identification of Network Applications. In PAM 2005, Boston, USA,
March 31-April 1, 2005.
[12] A. W. Moore and D. Zuev. Internet Traffic Classification Using
Bayesian Analysis Techniques. In SIGMETRIC’05, Banff, Canada,
June 6-10, 2005.
[13] V. Paxson. Empirically-Derived Analytic Models of Wide-Area TCP
Connections. IEEE/ACM Transactions on Networking, 2(4):316–336,
August 1998.
[14] M. Roughan, S. Sen, O. Spatscheck, and N. Duffield.
Class-of-Service Mapping for QoS: A Statistical Signature-based
Approach to IP Traffic Classification. In IMC’04, Taormina, Italy,
October 25-27, 2004.
[15] S. Sen, O. Spatscheck, and D. Wang. Accurate, Scalable In-Network
Identification of P2P Traffic Using Application Signatures. In
WWW2005, New York, USA, May 17-22, 2004.
[16] I. H. Witten and E. Frank. (2005) Data Mining: Pratical Machine
Learning Tools and Techniques. Morgan Kaufmann, San Francisco,
2nd edition, 2005.
[17] S. Zander, T. Nguyen, and G. Armitage. Automated Traffic
Classification and Application Identification using Machine
Learning. In LCN’05, Sydney, Australia, Nov 15-17, 2005.
