You are on page 1of 8

International Journal of Engineering and Techniques - Volume 3 Issue 6, Nov - Dec 2017

RESEARCH ARTICLE OPEN ACCESS

Identifying Telecommunication Deception using


Neural Networks through Data Mining
Gurunadham R
Assistant Professor, Dept of CSE
Sree Dattha Institute of Engineering and Science, Telangana

Abstract:
Neural computing refers to a pattern recognition methodology for machine learning. The resulting model from neural
computing is often called an artificial neural network (ANN) or a neural network. Neural networks have been used in many
business applications for pattern recognition, forecasting, prediction and classification. Neural network computing is a key
component for any data mining tool kit. Applications of neural network based data mining tools are abound in finance,
marketing, manufacturing, information systems and so on. This essay discusses in detail the role of artificial neural networks in
prevention of telecommunication fraud.
Keywords--- Artificial iintelligence, Artificial neural networks (ANNs), Coral systems, Data Mining, Expert System, Falcon
(Or Credit Card Issuers), Fraud Management Systems, Ghosting, Knowledge Base System, Rule Based System,
Telecommunication fraud, Thresholding Systems.

1 AN OVERVIEW OF At the beginning of the twenty first century,


TELECOMMUNICATION FRAUD the convergence of computing and
communication technologies has altered
THE telecommunication industry has considerably the way in which industrialized
expanded dramatically in the last few years communities function. It has created unfold
with the development of fordable mobile benefits for education, delivery of health
phone technology (Pieprzyk J, Ghodosi H services, recreation and commerce and
and Dawson E, 2007, pp 446-447). With the changed considerably the nature of modern
increasing number of mobile phone workplaces and patterns of employment.
subscribers, global mobile phone fraud is Telecommunication fraud can be simply
also set to rise. It is a worldwide problem described as any activity by which
with substantial annual revenue losses of telecommunications service is obtained
many companies. Telecommunication fraud without intention of paying. This kind of
which is the focus is appealing particularly fraud has certain characteristics that make it
to fraudsters as calling from the mobile particularly attractive to fraudsters. The
terminal is not bound to a physical location main one is that the danger of localization is
and it is easy to get a subscription. This small. This is because all actions are
provides a means for illegal high profit performed from a distance which in
business for fraudsters requiring minimal conjunction with the mess topology and the
investment and relatively low risk of getting size of network makes the process of
caught. Telecommunication fraud is defined localization time consuming and expensive.
as the unauthorized use, tampering or Additionally no particularly sophisticated
manipulation of a mobile phone or service. equipment is needed if one is needed at all.
The simple knowledge of an access code,
which can be acquired even with methods of

ISSN: 2395-1303 http://www.ijetjournal.org Page 266


International Journal of Engineering and Techniques - Volume 3 Issue 6, Nov - Dec 2017

social engineering, makes the analysis. A rule based system is a set of


implementation of fraud feasible. Finally the rules that take into account the normal
product of telecommunication fraud, a calling hours, the called destinations as well
phone call is directly convertible to money. as the normal duration of the call and etc. 3)
Adaptive flexible techniques using advanced
2. TELECOMMUNICATION FRAUD data analysis like artificial neural networks
DETECTION (ANNs). Fed with raw data, a neural
network can quickly learn to pick up
Fraud is a multi billions problem around the patterns of unusual variations that may
globe. The problem with telecommunication suggest instances of fraud on a particular
fraud is the huge loss of revenue and it can account (Liatsis P, 2002, pp 474-475).
affect the credibility and performance of
telecommunication companies. The most 3 TYPES OF TELECOMMUNICATION
difficult problem that faces the industry is FRAUD
the fact that fraud is dynamic. This means The telecom industry suffers major
that whenever fraudster’s feel that they will losses due to fraud (Prasad S K, Routray S
be detected they find other ways to and Khurana R, 2009, pp 259-260; Żytkow J
circumvent security measures. M and Rauch J, 1999, pp 251). There are
Telecommunication fraud also involves the many different types of telecommunications
theft of services and deliberate abuse of fraud and these can occur at various levels.
voice and data networks. In such cases the The two most types of fraud are subscription
perpetrator’s intention is to completely fraud and superimposed fraud.
avoid or at least reduce the charges for using 3.1 Subscription Fraud
the services. Over the years, fraud has In subscription fraud, fraudsters obtain an
increased to the extent that losses to account without intention to pay the bill.
telephone companies are measured in terms This is thus at the level of a phone number,
of billions of American dollars. Fraud all transactions from this number will be
negatively impacts on the telephone fraudulent. In such cases abnormal usage
company in 4 ways such as financially, occurs throughout the active period of the
marketing, customer relations and account. The account is usually used for call
shareholder perceptions. selling or intensive self usage.
3.2 Superimposed Fraud
There are various techniques available for In Superimposed fraud, fraudsters take over
managing and detecting telephone fraud a legitimate account. In such cases the
these include: 1) Manual review of data, the abnormal usage is superimposed upon the
problem with this technique is the fact that normal usage of the legitimate customers.
there are too many data records for a team to There are several ways to carry out
filter the fraudulent data. Typically a superimposed fraud, including mobile phone
telecom company will have in order of 1 cloning and obtaining calling card
million or more records of telephone calls authorization details. Examples of such
generated by their customers for a single cases include cellular cloning, calling card
month within a specific region. As a result theft and cellular handset theft.
this is a time consuming and laborious Superimposed fraud will generally occur at
technique for detecting fraud. 2) the level of individual calls; the fraudulent
Conventional analysis using a fixed rule calls will be mixed in with the justified ones.
based expert system together with statistical Other types of telecommunications fraud

ISSN: 2395-1303 http://www.ijetjournal.org Page 267


International Journal of Engineering and Techniques - Volume 3 Issue 6, Nov - Dec 2017

include ghosting (technology that tricks the infrastructure. The fraud techniques are
network in order to obtain free calls) and considered because basic ideas remain
insider fraud where telecommunication despite the underlying technology.
company employees sell information to Deceptions in telecommunications include
criminals that can be explained for subscription frauds where the cheater
fraudulent gain. accesses the services without being
subscribed. User can also suffer line or
4. METHODS OF identity theft being charged for services used
TELECOMMUNICATION FRAUD by others. Telecommunication operators can
There are two methods by which oversee users that exceed their download
telecommunication fraud can be quote and rate performing illegal service
implemented. They are discussed below in redistribution, sometimes for an economic
detail. profit. Finally cloning or unauthorized
4.1 Theft of communication Services: access to services may lead to compromising
The phone phreakers of 3 decades ago set a privacy.
precedent for what is becomes a major Anyway the most common types of fraud on
criminal industry (Wall D S, 2001, pp 30; telecommunications are subscription fraud
Broadhurst R G and Grabosky P N, 2005, pp and identity theft. After that voice mail fraud
32). The market for stolen communications and calling card fraud prevail. The analysis
services is now large. There are those who of different fraud techniques points out that
simply seek to avoid or to obtain a discount the tendency is a convergence of the fraud
on the cost of a telephone call whilst there which increases the complexity of its
are others such as illegal immigrants who detection. Fraud management systems have
are unable to acquire legitimate information proved to be a suitable tool to detect fraud in
services without disclosing their identity and different networks with diverse techniques
their status. such as self organizing maps (SOM),
4.2 Communications in Furtherance of general data mining, rule based systems
criminal conspiracies: profiling through Artificial intelligence
Just as legitimate organizations in the techniques like neural networks or decision
private and public sectors rely upon trees based on the hierarchical regime
information systems for communications switching models, Bayesian networks, fuzzy
and record keeping so too are the activities rules or other data mining techniques. There
of criminal organizations enhanced by also exist works on how to discover new
technology. rules to detect fraud in telecommunications
and on the privacy concerns of applying
detection techniques to user’s data.

5 APPROACHES TO FACE FRAUD Fraud detection can also be done at 2 levels


call or behavior and with two different
Dealing with the fraud is a very complex approaches user profile or signature based
task mainly due to its transversal nature to (Perner P, 2006, pp 535). Most of the
the operator’s structure (Samarati P, 2010, techniques use the CDR data to create a user
pp 201). Traditional fraud techniques are profile and to detect anomalies based on
evolving and adapting to the new network these profiles. The mined large amounts of

ISSN: 2395-1303 http://www.ijetjournal.org Page 268


International Journal of Engineering and Techniques - Volume 3 Issue 6, Nov - Dec 2017

CDR have in order to find patterns and 6 TELECOMMUNICATION FRAUD


scenarios of normal usage and typical fraud DETECTION USING NEURAL
situations. These scenarios were then used to NETWORKS
configure monitors that observe the user Artificial intelligence technologies have
behavior with relation to that type of fraud. been a part of fraud detection systems in
These monitors are then combined in a several industries for half dozen years
neural network which raises an alarm when (Liebowitz J and Prerau D S, 1995, pp 177).
sufficient support of fraud exists. This type Hecht Nielson Neuro-computers of San
of system can be classified in a rule based Diego, California provide a neural network
approach since it relies in the triggering of based fraud detection system called Falcon
certain rules due to abnormal usage. The or credit card issuers. In the
rule based system has the drawback of telecommunication area, a Longmont,
requiring expensive management of rules. Colorado company called Coral Systems
Rules need to be precise (avoid false offers a knowledge based systems for
positive alarms) and constantly evolving cellular fraud detection. This system uses
(detect new scenarios) which result in very expert fraud rules to sport potentially
time consuming programming. fraudulent calls outside the normal pattern of
The most common and best succeeded a caller’s activity. A caller who makes calls
methods for fraud analysis are signature in a rapid succession is probably the Victim
based. These methods detect the fraud based of a clone phone, a common practice among
on deviation detection by comparing the cellular thieves for stealing and using a
recent activity with the user behavior data legitimate cellular amount. If a caller who
which is expressed through the user makes 45 calls a day on average suddenly
signature. The work can be adapted and makes 50, no fraud alert is sounded. But if a
extended by reformulating the notion of different caller who may make only a few
signature and by introducing the notion of calls a week suddenly makes 50 in a day the
statistical based distances to detect account is flagged for fraud investigation.
anomalies. Furthermore the computational The system pulls information from carrier
cost can be reduced by using simple switches and looks for deviations from a
statistical functions avoiding processing caller’s normal patterns of phone use. The
costly histograms. A clear problem with a system can be tuned to produce alerts for
histogram approach is that discretization any call which looks suspicious or can be
intervals or bucket must be clean and what is tuned to provide as few false positives as
right for one customer may be wrong for possible, which raises the fraud rate among
another. flagged calls to 90-95% but may miss many
Other approaches have also been widely frauds. The systems have been out since
applied to fraud analysis like neural 1992 and according to Coral systems are in
networks. Another applied technique is link use by 5 customers. While this system does
analysis. Here the client links are updates not currently use neural networks Coral is
over time establishing a graph of called investigating them for future use.
communities of interest that can easily
reveal networks of fraudster’s. These
methods are based on the observation that 7 .FRAUD DETECTION SYSTEM
fraudsters seldom change their calling habits USING ARTIFICIAL NEURAL
but are often linked to other fraudsters. NETWORKS

ISSN: 2395-1303 http://www.ijetjournal.org Page 269


International Journal of Engineering and Techniques - Volume 3 Issue 6, Nov - Dec 2017

consider the near real time nature of


emerging frauds and the need to close down
potential high cost fraud quickly. Globally
the usual experience is that fraud
management system pays for their
investment in very short times provided they
are directed at the principal sources of fraud.

Data sources for fraud detection systems are


principally Call Detail Record (CDR) based,
deriving usually from the host billing
system. For a more real time data source the
PSTN itself is the usual source. In the
narrowband world this will mean monitoring
Number 7 signaling messages at strategic
Fig.1. Fraud Detection System using ANN points in the network. Network topology
Technology then becomes an issue. In a network with
An overview of the Fraud Detection System widespread use of Signal Transfer Point
developed by using Artificial Neural (STP) working, signaling is concentrated in
Network Technology is shown in the below a relatively few points and monitoring may
figure: be economically infeasible so a more
The process begins with gathering historical focused approach is called for. The usual
data on fraudulent and non fraudulent calls. axiom is to follow the money. Therefore
This data is preprocessed to make it suitable monitoring international routes would be a
for neural network learning. Next the neural priority. One could also consider monitoring
network is trained using the preprocessed the Intelligent Network platform as this
data to build a model which incorporates usually provides a signaling concentration
numerous patterns of fraudulent behavior. point and may also deal with high value
The model is applied to incoming business number translation services such as
where it adaptively learns new patterns of Premium Rate. As voice over IP services
fraud improving its model as the types of roll out, fraud detection system
fraud evolves. manufacturers are turning their attention to
the specific needs of extracting or creating
Fraud detection mechanisms may be home CDR’s from telephony services or other
brew, proprietary or a mix of both which is network components. The fraud detection
probably the healthiest approach (Oodan A mechanism consists of a rule based system
P, 2003, pp 493). A web search for telecom for detecting fraud and a self learning
fraud management will result in a system that makes this system adaptive. A
bewildering number of hits. Simple rule based tool was a white box approach
thresholding systems work well when that allowed detecting the frauds with low
integrated into business as usual processes rate or false alarms (Prasad R, 2001, pp
and could well account for 80% or so of the 265).
fraud management load. For more complex
situations where product interaction is More specifically the rule based system
taking place, some form of correlation is contains an ontology repository which is
required using neural or artificial practically the knowledge base of the system
intelligence techniques. One should also

ISSN: 2395-1303 http://www.ijetjournal.org Page 270


International Journal of Engineering and Techniques - Volume 3 Issue 6, Nov - Dec 2017

and which stores all the domain and fraud largest telecom operator was losing over
specific knowledge that is required by the United States $37 million per year to fraud.
system (Filipe J and Obaidat M S, 2008, pp Subscription fraud in which a customer
112-113). The actual fraud detection either provides fraudulent details or gives
methods and techniques are stored in the valid details and then disappears was the
Semantic Rules Repository in the form of if- company’s biggest cause of revenue
then-else rules that reason over the leakage. By the time the telecom provider is
knowledge contained in the ontology alerted about the fraud, the fraudster has
repository. The two repositories are already moved to other target victims. Other
interrelated as the representation of rules types of fraud include phone card
depends on the contents of the ontology. manipulation which involves tampering and
cloning of phone cards. In a clip on fraud a
Finally the self learning system provides the fraudster clips on to customers telephone
overall system with the capability to learn lines and then sells calls to overseas
new rules about fraud from the submission destination for a fraction of normal rates.
of fraudulent and non fraudulent domain Minotaur, developed by Neural
specific data and to automatically detect Technologies was implemented to prevent
irregular observations in the data thus fraud. Minotaur uses a hybrid mixture of
providing a feedback mechanism for intelligent systems and traditional
enriching and updating the repository of computing techniques to provide customer
rules. The self learning system integrates subscription and real time call monitoring
suitable algorithms for statistical data fraud detection. It processes data from
analysis and data mining tasks that enable it numerous fields such as event data records
to update, optimize and extend existing (switch/CDR, SS#7, IPDRs,
fraud detection rules by analyzing submitted PIN/authentication) and customer data
data. (billing and payment, point of sale,
provisioning) using a multi-stream analysis
The fraud detection area is an active area of capacity. Frauds are detected on several
development for neural networks in levels such as on an individual basis using
telecommunications. Many of the most specific knowledge about the subscriber’s
successful systems are hybrid systems which usage and on a global basis using generic
takes advantage of the relative strengths of knowledge about subscriber usage and
several AI technologies. Given the payoffs known fraud patterns. The neural capability
involved it is an application which should of Minotaur means it learns from experience
come into routine use in the years ahead. making use of adaptive feedback to keep up

to date with changing fraud patterns. In the


8. HOW DOES NEURAL NETWORK first three months of installation of this
HELP IN REDUCING neural network based software: The average
TELECOMMUNICATIONS FRAUD? fraud loss per case was reduced by 40%. and
The Forum of International Irregular The detection time was reduced by 83%..
Network Access (FIINA) estimates that The combination of neural, rule based and
telecommunication fraud results in a loss of case based technologies provide a fraud
United States $55 billion per year worldwide detection rate superior to that of
(Turban, 2008, pp 349). South Africa’s conventional systems. Furthermore the multi

ISSN: 2395-1303 http://www.ijetjournal.org Page 271


International Journal of Engineering and Techniques - Volume 3 Issue 6, Nov - Dec 2017

stream analysis capability makes it the next century. As a result Artificial


extremely accurate. Neural Network is a better method for
detecting telephone fraud, due to its inherent
9. CONCLUSIONS ability to adapt as well as its speed and
efficiency.
The theft of telecommunication services has
been one of the most enduring types of 10. REFERENCES
telecommunications crime which has been
evident since the beginning of telephone [1] Pieprzyk J, Ghodosi H and Dawson E
systems (Grabosky P N and Smith R G, (2007), Information security and privacy:
2009, pp 84). Fraud detection for mobile 12th Australasian conference, ACISP 2007,
telecommunications is a relatively recent Townsville, Australia, July 2-4, 2007:
area of research. Other works in the area of proceedings, Springer, Germany, pp 446-
fraud detection in mobile 447.
telecommunications are based on data [2] Liatsis P (2002), Recent trends in
mining approaches (Althoff K D, 2008, pp multimedia information processing:
563).Due to its characteristics this type of proceedings of the 9th International
fraud requires real time and individualized Workshop on Systems, Signals and Image
customer analysis. Each technological Processing, World Scientific Publishing,
development designed to thwart criminal London, pp 474-475.
endeavors has been quickly followed by the [3] Prasad S K, Routray S and Khurana R
creation of a new form of crime designed to (2009), Information Systems, Technology
exploit new security. These few directions and Management: Third International
of future policy may assist in ensuring that Conference, ICISTM 2009, Ghaziabad,
the full potential of global India, March 12-13, 2009, Proceedings,
telecommunications developments will be Springer, Germany, pp 259-260.
realized while at the same time providing [4] Żytkow J M and Rauch J (1999),
both service providers and users with some Principles of data mining and knowledge
expectations that their property rights will be discovery: Third European Conference,
respected. PKDD'99, Prague, Czech Republic,
Fraud detection will continue their September 15-18, 1999 : proceedings,
accelerated use of neural network based Springer, USA, pp 251.
systems. Many of the laboratory techniques [5] Wall D S (2001), Crime and the Internet,
for using neural networks in metro burst Routledge, London, pp 30.
communications, satellite network
management and traffic control will come
on line. In short the neural networks will [6] Broadhurst R G and Grabosky P N
become an increasing presence in major (2005), Cyber-crime: the challenge in Asia,
aspects of telecommunication networks Hong Kong University Press, Hong Kong,
improving efficiency, adapting to changing pp 32.
calling patterns, and providing better [7] Samarati P (2010), Information Security
information about the use of networks. Theory and Practices: Security and Privacy
Neural networks a technology which has of Pervasive Systems and Smart Devices:
been used in telephony since the early 1960s 4th IFIP WG 11.2 International Workshop,
is beginning to make it presence felt in WISTP 2010, Passau, Germany, April 12-
designing in telecommunications network of 14, 2010, Proceedings, Springer, USA, pp
201.

ISSN: 2395-1303 http://www.ijetjournal.org Page 272


International Journal of Engineering and Techniques - Volume 3 Issue 6, Nov - Dec 2017

[8] Perner P (2006), Advances in data


mining: applications in medicine, web
mining, marketing, image and signal mining
: 6th Industrial Conference on Data Mining,
ICDM 2006, Leipzig, Germany, July 14-15,
2006 : proceedings, Springer, Germany, pp
535.
[9] Liebowitz J and Prerau D S (1995),
worldwide intelligent systems: approaches
to telecommunications and network
management, IOS Press, Netherlands, pp
177.
[10].OodanA P (2003), Telecommunications
quality of service management: from legacy
to emerging services, The Institution of
Engineering and Technology, London, pp
493.
[11] Filipe J and Obadiah M S (2008), E-
business and telecommunications: 4th
International Conference, ICETE 2007,
Barcelona, Spain, July 28-31, 2007, revised
selected papers, Springer Verlag, Germany,
pp 112-113.
[12] Prasad R (2001), towards a global 3G
system: advanced mobile communications in
Europe, Artech House, USA, pp 265.
[13] Turban (2008), Decision Support and
Business Intelligence Systems, 8/E, Dorling
Kindersley, New Delhi, pp 349.
[14] Grabosky P N and Smith R G (2009),
Crime in the digital age: controlling
Telecommunications and cyberspace
illegalities, Transaction Publishers, USA, pp
84.
[15] Althoff K D (2008), Advances in case-
based reasoning: 9th European conference,
ECCBR 2008, Trier, Germany, September
1-4, 2008 ; proceedings, Springer, USA, pp
563.

ISSN: 2395-1303 http://www.ijetjournal.org Page 273

You might also like