GLOBAL SYSTEM FOR MOBILE COMMUNICATION (GSM) SUBSCRIPTION FRAUD DETECTION SYSTEM USING ARTIFICIAL NEURAL NETWORK TECHNIQUE

ABSTRACT

This project is concerned with GSM subscription fraud detection system using artificial neural network technique. Fraud is a multi-billion problem around the globe with huge loss of revenue. Fraud can affect the credibility and performance of telecommunication companies. The most difficult problem that faces the industry is the fact that fraud is dynamic, which  means that whenever fraudsters feel that they will be detected, they device other ways to circumvent security measures. In such cases, the perpetrators intention is to completely avoid or at least reduce the charges for using the services. Subscription fraud is one of the major types of telecommunication fraud in which a customer obtain an account without intention to pay the bill. Thus at the level of a phone number, all transactions from the number will be fraudulent. In such cases abnormal usage occurs throughout the active period of the account; which is usually used for call selling or intensive self usage. This provides a means for illegal high profit business for fraudsters requiring minimal investment and relatively low risk of getting caught. A system to prevent subscription fraud in GSM telecommunications with high impact on long distance carriers is proposed to detect fraud. The system employs adaptive flexible techniques using advanced data analysis like Artificial Neural Networks (ANN). Fed with raw data, a neural network can quickly learn to pick up patterns of unusual variations that may suggest instances of fraud on a particular account. A total of 158 data samples were collected, trained and tested using a model that allows identifying potential fraudulent customers at the time of subscription. The result shows that 80% of the prediction accuracy has been obtained. From the result produced, artificial neural network has a potential to be used for detecting subscription fraud in telecommunication.

CHAPTER ONE

INTRODUCTION

1.1 BACKGROUND OF THE STUDY

            Following the exponential growth in the telecommunications sector in the end of past century, the telecommunications operators face a new challenge: fraud. It is not only a risk, but a highly organized global business, that affects operators all over the world. In order to realize the severity of this problem, Communication Fraud Control Association (CFCA) published some statistics stating that the annual global fraud losses in the telecoms sector are now between US$54 Billion and $60 Billion, an increase of 52% since 2003[6].Global System for Mobile (GSM) fraud has identify itself as the single biggest cause of revenue loss for telecom carriers with the increasing number of mobile phone users, global mobile phone fraud is also set to rise. [29]distinguished different fraud scenarios: subscription fraud, dial through fraud, free phone fraud, handset theft and roaming fraud. The Cambridge Advanced learners Dictionary defined fraud as “the crime of obtaining money by deceiving people”, while the concise Oxford Dictionary defines it as a “criminal deception; the use of false representations to gain an unjust advantage”. GSM fraud can be simplified described as any activity by which service is obtained without intention of paying. Using this definition fraud can only be detected once it has occurred. In subscription fraud, the typical behaviour of fraudsters is to abuse service by making significant usage of telecom services (for example, calling, messaging, internet, etc) before the bill is served. Customers applications are sometimes rejected by the company at the time of application if they find that it is risky to entertain customers who are likely to hold bad dept. Estevez et al (2006) propose neural network model to detect subscription fraud at the time of  application.

1.2 STATEMENT OF PROBLEM

Currently, due to the development of new technologies, traditional fraudulent activities, such as money laundering, have been joined by new kind of fraud like GSM fraud and computer intrusion. Fraud is increasing dramatically each year resulting in loss of a large amount of money worldwide.

During the research, some problems have been identified as a likely cause of increase in subscription fraud. They are:-

1. The use of GSM lines without proper registration of SIM card.
2. Call roaming, that is making calls outside home system.
3. Signing up GSM telecom service using false or stolen identification.
4. No standard fraud detection system to checkmate fraudsters.

1.3 OBJECTIVES OF THE STUDY

The main objective of this work is to detect fraud occurrence in GSM network. The specific objective includes the following:-

1. To investigate and identify fraud inherent in GSM telecom.
2. To develop an architecture for the detection of GSM telecom fraud using neural network.
3. To develop a piece of software for the detection of GSM telecom fraud.
4. To evaluate the functionality of the developed system.

   1.4 SCOPE AND LIMITATION OF THE STUDY

This project work is to develop a GSM Subscription Fraud Detection System. This fraud detection system will focus on detecting as many subscription fraudsters as possible and neural network technique is used to detect the subscription fraudsters.

Some of the constraints encountered during this project design include the following:

• Financial Constraints: The design was achieved but not without some financial involvements. One had to pay for the computer time. Also the typing and planning of the work has its own financial involvements.  However, to solve the problems I solicited fund from guardians and relations.
• High programming Technique: The programming aspect of this project posed a lot of problematic bugs that took me some days to solve. Also other technical problem, which requires semantic and syntactic approaches where encountered as well. In seeking for the solution to these problems, I acquired more knowledge from well –versed textbooks and programmes.
• The epileptic nature of power supply cannot be overlooked.  

1.5 SIGNIFICANCE OF THE STUDY

If fraud is properly handled using artificial neural network, there are more benefits to both the subscriber and service provider. The benefits are:-

1. To help prevent revenue loss
2. To detect untrustworthy dealers
3. To reduce widespread costs by subscription fraud
4. To identify fraudsters, when using a service that is not properly registered.

1.6 DEFINITION OF TERMS

SIM: Subscriber Identity Module; A smart card containing the telephone number of the subscriber, encoded network identification details, the PIN and other user data such as the phone book. A user’s SIM card can be moved from phone to phone as it contains all the key information required to activate the phone.

Telecommunication: Are devices and systems that transmit electronic or optical signals across long distances. Telecommunication enables people around the world to contact one another to access information instantly, and to communicate from remote areas.

Computer Network: It is a system used to connect two or more computers using a communication link.

Subscription Fraud: Is defined as a use of telecommunication products or services without intension to pay (wikipedia- CFCA’S, 2011 worldwide telecom fraud survey).

GSM: Global system for mobile communication is a time division multiple access (TDMA) based wireless network technology developed in Europe that is used throughout most of the world. GSM phones makes use of SIM card to identify the users account; which also makes it ease for the user to quickly move their phone number from one GSM phone to another by simply moving SIM card.

TDMA: time division multiple access is a multiplexing method that divides network connections into time slices, where each device on the TDMA network connection gets one or more time slice during which it can transmit or receive data.

ANN: an artificial neural network is a computational model based on the structure and functions of biological neural networks. Information that flows through the network affects the structure of the ANN because a neural network changes or learns, in a sense based on that input and output.