A Web-based Interactive Data Visualization System for Outlier Subspace
Analysis. Dong Liu, Qigang Gao. Computer Science. Dalhousie University.
Halifax, NS ...
A Web-based Interactive Data Visualization System for Outlier Subspace Analysis Dong Liu, Qigang Gao Computer Science Dalhousie University Halifax, NS, B3H 1W5 Canada
[email protected] [email protected]
Hai Wang Sobey School of Business Saint Mary’s University Halifax, NS, B3H 3C3 Canada
[email protected]
Ji Zhang Mathematics & Computing University of Southern Queensland Toowoomba, QLD, 4350 Australia
[email protected]
log data, cancers in medical data, or simply some errors or
Abstract
noises caused by human mistakes or sensor damage, etc Detecting outliers from high-dimensional data is a
[11, 12, 13]. Outliers should be treated differently in
challenge task since outliers mainly reside in various low-
different situations, such as errors and noises outliers
dimensional subspaces of the data. To tackle this
should be removed, and intrusion and cancer outliers are
challenge, subspace analysis based outlier detection
targets and should be detected for analysis and event
approach has been proposed recently. Detecting outlying
prevention. In other situation, outliers must be detected
subspaces in which a given data point is an outlier
and classified properly.
facilitates a better characterization process for detecting
Traditional outlier detection methods are mainly been
outliers for high-dimensional data stream, and make
designed using whole dimensionality analysis approach.
outlier mining for large high-dimensional data set to be
They work well for low-dimensional data sets. However,
more manageable.
In this paper, to facilitate outlier
nowadays more and more real applications are involved in
subspaces analysis from human perception perspectives in
high-dimensional data. Detecting outlier from high-
supporting the development of efficient solutions for
dimensional data is a challenging task, in that traditional
high-dimensional
web-based
methods become infeasible for high-dimensional data due
interactive data visualization system, which can display
to the Curse of Dimensionality phenomena, in that the
various low-dimensional outlier subspaces to allow users
outliers hidden in low-dimensional subsets of the data will
to observe and analyze the distributions of outliers. The
be disappeared as the dimensionality is increased for
proposed visualization tool can help the developers of
using whole dimensionality analysis methods [2]. The
outlier detection applications to directly examine the
new strategy to deal with high-dimensional data is to
distributions of outliers in various low-dimensional
detect outliers for possible lower dimensional subspaces
subspaces to validate their experiment results.
of the high-dimensional data, such as introduced in [1].
data,
we
propose
a
The idea is to convert the issue of outlier detection in the 1
Introduction
high-dimensional data space into the issue of detecting low-dimensional outlying subspaces since exhaustive
Outliers in a database or data stream are the data
search all subspaces in high-dimensional data space is not
objects that are grossly different from or inconsistent with
tractable. In this paper, we propose a data visualization
the rest of the data, which reflect abnormal behaviours in
system to facilitate analysis and solution development for
the real world. Outliers may stand for toxin spills in
projected outlier subspace finding and gain insight by
chemical sensor data, the network intrusions in network
allowing the developers/users to observe the data
distributions for various low-dimensional outlier subspace
= {,o∈O and S is the outlying subspace set of o},
of the data.
where O denotesset of outliers detected.
Visualization has been proved to be a useful tool for
The visualization system aims to help users to
data analysis. With development of computer hardware
examine the detected outlying subspaces for high-
and software, visualization techniques can use computer
dimensional data set. Users are allowed to adjust the
graphics
in
parameters of the outlier detection algorithms and
understanding of complex, often massive representations
visualize the intermediate detection results. A set of
of data. There are a number of visualization tools
visualization tools is designed for supporting human
available, such as SequoiaView [3], GGobi [6], OpenViz
exploration on projected outlier subspace analysis.
to
create
visual
images
which
aid
[7], VisuMap [8] and ADVIZOR [9]. Some tools are webbased systems for the continence of accessing the tool for
2.1
System Architecture
broad user groups, such as Manyeyes [4] and Drillet [5]. However, there is no data visualization system for directly
The architecture of the visualization system is
analyzing projected outlier subspaces. In this paper, we
illustrated in Figure 1.
present a visualization system for outlier subspace
include both the original high-dimensional data set and
analysis in that the features and interface tools are special
the outlier detection results after data pre-processing
designed for effectively supporting human to observe and
which includes standard steps of data cleaning, data
explore large volume high-dimensional data for gaining
transformation and data normalization. Data cleaning is to
insight on outlier detection on such complex data sets.
remove
incorrect
The data to be displayed can
records
in
the
dataset.
Data
transformation is to correct inconsistent data format and 2
System Design and Implementation
convert continuous data attribute values into a finite set of intervals with minimal loss of information. In data
The proposed visualization system is designed for
normalization, we will find out the minimum and
supporting outlier analysis on high-dimensional data in
maximum value for each dimension and convert value
that human perception can play a role for gaining insight
between 0 and 1.
on outlier subspaces, which is based on the concept of “Stream Projected Outlier Detector (SPOT)” [1]. In SPOT system, the problem of detecting projected outliers from high-dimensional data streams is formulated as follows. Given a data streamD with a potentially unbounded size of ϕ-dimensional data points, each data point pi = {pi1, pi2, . . . , pi'} in D will be labeled as either a projected outlier or a regular data point. If pi is a projected outlier, its associated outlying subspace(s) will be given as well. The results to be returned will be a set of projected
Figure 1 System Architecture
outliers and their associated outlying subspace(s) to indicate the context where these projected outliers exist.
For the prepared high-dimensional data, one data
The results, denoted by A, can be formally expressed as A
point may be considered as outlier in many subspaces,
therefore the outlier detection result may be very large. In
dimensional subspaces. Below is a sample of the first two
order to handle large size of outlier detection results, the
detected outlying subspaces in the file.
system to use a database to store the datasets and the
Outlierness Threshold: 3
information of outlying subspaces. After data preparation
*****************************************
stage, both the datasets and the outliers are stored into two
Top outlier: data #1
tables in the database. By doing so, the database server
In subspace: 11
can quickly retrieve the selected data for feeding into the
Cell index: 1
visualization system for display. With the prepared data
Outlier-ness: 3.3557
sets, the user should be able to access the system through
Top outlier: data #2
internet with a web browser. The system allows the user
In subspace: 1 6
to select different subspaces and – views to display.
Cell index: 15 6
According to user’s subspace selection, the system will
Outlier-ness: 3.57143
connect to the database server with JDBC and send
... ...
queries to database server. The retrieved data and outlier information for the selected subspaces will be transmitted
Field
Type
to client machine over internet and displayed in user’s web browser.
Description Primary Key. Row number of
linenumber
int(11)
data.
The database and web application services are at
valume1
double
Attribute 1
server side. On the client side, user can access the web
valum2
double
Attribute 2
services and visualize data and outliers for the selected
...
...
...
subspaces from the web browser. The system also allows
valume15
double
Attribute 15
the user to visualize different datasets by reading data file Table 1 Schema of Data in Database
name specified by the user from user’s local machine. The system is implemented in Java. The client machine needs to install J2SE 5 and Java 3D 1.5 or higher version to run the system.
2.2
Synthetic Datasets In the experiments, both synthetic data and real data
sets are used. The synthetic data is generated randomly by a high-dimensional data generator
used in SPOT
research [1]. The nature of the data is close to real-life data.
It
exhibits
different
data
characteristics
Field
Type
Description Primary Key and identify each
id
int(11)
outlying subspace.
linenumber
int(11)
Row number of data.
dimension1
int(11)
Attribute 1 of outlying subspace.
dimension2
int(11)
Attribute 2 of outlying subspace.
dimension3
int(11)
Attribute 3 of outlying subspace.
outlierness
double
Outlierness of outlier.
Table 2 Schema of Outlier Information in Database
in
projections of different subsets of features. It consists of
Since the outlier detection result contains only
15 attributes and 10,000 lines of data. The outlier
outlying subspaces of 1, 2 and 3 dimensional subspaces.
detection result directly from SPOT method [1] consists
The corresponding data tables and outlier table are created
of 426,513 outliers from one dimensional to three
in the database. The detailed schema of the data table is
illustrated in Table 1. The detailed schema of the outlier
cases for both synthetic datasets and KDD 1999 network
table is given in Table 2. The attribute values of outlying
log data. The visualization system can help to answer
subspaces are sorted in ascending order. For one-
questions on the outlier detection. For examples,
dimensional outlying subspaces, the values of dimension2 and
dimension3
dimensional
are
outlying
NULL.
Similarly,
subspaces,
the
for
attribute
1. In a two-dimensional subspace of the synthetic
two-
datasets, find out whether a selected particular outlier data
of
point is also an outlier in other two-dimensional
dimension3 is NULL. For three dimensional outlying subspaces, values of all dimensions are not NULL.
subspaces. 2. What distribution of “smurf” network attacks is in KDD 1999 data?
2.3
Real-life Datasets
Case 1: In a two-dimensional subspace, find out whether a selected outlier data point is also considered as
The experiments also include real-life data sets, i.e.
an outlier in other two-dimensional subspaces.
the KDD Cup 1999 data [10], which is a log connection
For answering this question, we visualize four two-
traffic data set from MIT/Lincoln-Lab. It contains
dimensional subspaces (as shown in Figure 2) which are
connections detail in its network such as the protocol-
(Dim4, Dim 6), (Dim3, Dim 6),( Dim 12, Dim 10) and
type, duration, service-use and many related information.
(Dim 2, Dim 4). When click one outlier (index #174) in
We use the first 5000 lines of the data from the corrected
subspace (Dim 4, Dim 6), then click the “Concurrent”
data with labels for our visualization. In the pre-
button in other two-dimensional subspace display
processing stage, we separate label information from
windows. We can easily observe that the outlier data point
datasets into a separated file. The label names are
(index #174) in (Dim4, Dim6) is also considered as
transformed into numbers. Each type of network intrusion
outlier in (Dim3, Dim 6) and (Dim 2, Dim 4). Moreover,
is mapping to one number. There are four types (shown in
we may change the outlierness threshold by moving slide
Table 3) of network intrusion labelled in the first 5000
bar in these two windows. We can get the outlierness
lines data. We use the number of outlier type as
value of data point (index 174) is 19.37 in both (Dim3,
outlierness value. In this way, we can visualize the
Dim 6) and (Dim 2, Dim 4).
distribution of different kind of network intrusion.
Case2: Visualize distribution of “smurf” network attack in KDD 1999 data. The example of visualizing the distribution of outliers in three-dimensional subspaces is shown in Figure 3. We may find out that the “smurf” network attacks are mainly resided closely in the marked area in the selected threedimensional subspace. Figure 4 is an example of use
Table 3 Label Mapping
concurrent display of two-dimensional subspaces. The system reports the selected outlier from the subspace in
3
Experiments and System Demonstration
The experiments are developed based on sample
left window is also marked as an outlier in the other subspace in the right window.
Figure 2 Case 1: Two-Dimensional Subspaces Concurrent Display
Figure 3 Case2: 3D Display
Figure 4 Case2: 2D Concurrent Display
4
Conclusion and Future Work [5] Drillet Visual Tool for interactive data analysis, http://drillet.appspot.com/. The proposed web-based visualization system can
help to observe subspaces of high-dimensional datasets
[6] Data Visulization system: GGobi, http://www.ggobi.org/.
interactively. - The system enables the user to evaluate performance of an outlier detection algorithm by visually
[7] Data Visulization system: OpenViz, http://www.avs.com/.
verifying the correctness of the results, and determining a proper parameter for better outlier detection results.
[8] Data Visulization system: VisuMap, http://www.visumap.net/.
Through visualizing datasets and their labelled results, user can gain insight visually on what real facts are about
[9] Data Visulization system: ADVIZOR, http://www.advizorsolutions.com/default.htm.
the data distribution nature and the outlier distribution. It is also useful for comparing the effectiveness of different
[10] KDD data source: http://kdd.ics.uci.edu/databases/kddcup99/kddcup99.html.
algorithms. The user may also adjust the values of different parameters of the algorithms for comparing the changes of performance. This system currently can visualize datasets and their labelled outlier information. It can interact with user and help to explore the datasets and outlier subspaces. In the future work, we may make the system to allow users to directly label outliers from selected subspaces. Users may also manually adjust outlierness value for selected outlier data points for observing sensitivity of the data. Moreover, the system may be integrated with different outlier detection algorithms such as the SPOT algorithm in [1].
5
References
[1] J. Zhang, Q. Gao and H. Wang. SPOT: A System for Detecting Projected Outliers from High-dimensional Data Streams. IEEE 24th International Conference on Data Engineering (ICDE’08), Cancun, Mexico, pp.1628-1631, 2008. [2] R. Bellman. Adaptive Control Processes: A Guided Tour. Princeton University Press, 1961. [3] Data Visualization system: http://www.win.tue.nl/sequoiaview/.
Sequoiaview,
[4] Data Visualization system: Manyeyes, http://manyeyes.alphaworks.ibm.com/manyeyes/.
[11] B. Aleskerov, E. Freisleben and B. Rao. Cardwatch: A Neural Network Based Database Mining System for Credit Card Fraud Detection. Computational Intelligence for Financial Engineering (CIFEr), 1997. [12] J. F. Costa. Reducing the Impact of Outliers in Ore Reserves Estimation. Mathematical Geology, 35(3), 2003. [13] J. Han and M. Kamber. Data Mining: Concepts and Techniques, 2nd ed. Morgan Kaufmann Publishers, 2006.