Classification of emails for Internet fraud detection and prevention through the application of artificial intelligence techniques.

Abstract

The social engineering threat known as Internet fraud is one among the forms of cybercrimethat cost huge financial loss to the global economy as recent data and statistics revealed thateven during the global Covid-19 pandemic Internet fraud continued to surge with many casesreported. Beyond financial losses, victims of Internet fraud in many instances suffers fromother problems like psychological trauma, job or business loss and so on. Internet fraud iscommitted by cyber criminals from different countries across the globe, but some forms ofInternet fraud have been identified and linked directly to specific countries. Advance fee fraudhas been particularly linked to Nigeria due to participation of Nigerians in such fraud and alsotracing the origin of many of such fraud to the country. Tackling internet fraud was amongthe reasons that necessitates Nigerian government to establish the Economic and FinancialCrimes Commission (EFCC), an agency responsible to fight Internet fraud, corruption andother financial crimes.This research addresses the detection and classification of Internet fraud through theapplication of artificial intelligence. A classifier is designed and implemented that classifiesincoming email into a category based on the content as either fraud or non-fraudulentapplying the Natural language process model of Bag of Words. The research focuses onAdvance fee fraud that originates from Nigeria, with part of the research data collected fromNigeria's law enforcement agency, the Economic and Financial Crime Commission (EFCC). Thisresearch is the first research work so far published that collect such dataset from thecommission to the best of my knowledge.The classifier is design and implemented in English language, and all the dataset used fortraining and testing are in English, therefore the classifier can be applicable and use by otherEnglish-speaking countries other than Nigeria. For countries that are non-English speaking,the Bag of word can be translated from English to their language and still be used for Internetfraud detection and classification.The classifier was implemented and experimented on using six machine learning algorithms;Decision Tree, Discriminant Analysis, Ensemble, Logic Regression, Nearest Neighbour andIVSupport Vector Machine to identify the one(s) that produce the highestclassification/detection rate compared to other similar published work.This research makes four main contributions to existing pool of academic research: (1) Focusand Identify specific Internet fraud directly linked to Nigeria. (2) Identify and generate uniquefeatures (Bag of Words) of Advance fee fraud that originate from Nigeria. This indeed is aunique contribution to research on Internet fraud. (3) Using artificial Intelligence technique,a classifier is design and implemented that successfully detect and classify fraudulent emailsfrom Non fraudulent ones based on their content. (4) Result of the classifier is presented, theClassifier produced the best result with Decision Tree algorithms achieving 100% detectionand classification accuracy using 7500 emails which consisted of 4500 fraudulent emails and3000 non fraudulent.

Date of Award	15 Jan 2021
Original language	English
Awarding Institution	University of Bedfordshire
Supervisor	Ali Mansour (Supervisor) & Vladimir Dyo (Second supervisor)