Constructing Subscription Fraud Detection Model Using Machine Learning Algorithms: the Case of Ethio Telecom
| dc.contributor.advisor | Beshah, Tibebe (PhD) | |
| dc.contributor.author | G/Tsadik, Hailemeskel | |
| dc.date.accessioned | 2021-03-16T06:29:43Z | |
| dc.date.accessioned | 2023-11-18T12:47:14Z | |
| dc.date.available | 2021-03-16T06:29:43Z | |
| dc.date.available | 2023-11-18T12:47:14Z | |
| dc.date.issued | 2021-01-05 | |
| dc.description.abstract | Nowadays, the advancement of telecom services are becoming an essential communication means for people’s day-to-day activities. However, this development provides some appearances that motivate fraudsters. Telecom fraud is a serious challenge in telecommunication industries. It is a threat for telecom companies to lose some percent of their annual revenue and to provide poor quality of services for their customers. Subscription fraud is one type of fraud in today’s telecom business. It is a common and major types of telecom frauds in which the usage category is in contradiction with the initial subscription type. The main objective of the fraudsters is to make money illegally or getting telecom services with the intention of not to pay for the service they used. The purpose of this study is to construct a model which uses machine learning Algorithms to detect subscription fraud calls by using Call Detail Records (CDR) data. The general approach used to perform this research was a quantitative laboratory experimental research method. Three classification techniques of machine learning algorithms have been applied; which are, Random Forest (RF), Support Vector Machine (SVM) and Artificial Neural Network (ANN) multilayer perceptron algorithms. WEKA data mining tool has been used to build a model for predicting fraudulent calls. The experimentation results of the work show that RF classifier performs better among the three algorithms with an accuracy of 99.46%. The major finding of this research is that, ten interesting factors used to identify fraudulent subscribers from legitimate ones. Some of the attributes such as, Subscribers total number of calls, number of unique called numbers, number of incoming calls, total international calls and ration of international total are important factors for domain expert practically practicing protecting the telecommunication frauds. There are misclassification results happened because of false positive and false negative. In telecom fraud detection, the cost of a false negative is more expensive than a false positive because a false positive can be classified correctly after further investigation, but a false negative means that the fraudster has managed to stay undetected and can continue committing fraud. Therefore further research needs to be done to reduce false negative in identifying subscription fraudsters. | en_US |
| dc.identifier.uri | http://etd.aau.edu.et/handle/12345678/25498 | |
| dc.language.iso | en | en_US |
| dc.publisher | Addis Ababa University | en_US |
| dc.subject | Subscription Fraud | en_US |
| dc.subject | Fraud Detection | en_US |
| dc.subject | Machine Learning | en_US |
| dc.subject | Support Vector Machine | en_US |
| dc.subject | Random Forest | en_US |
| dc.subject | and Artificial Neural Network | en_US |
| dc.title | Constructing Subscription Fraud Detection Model Using Machine Learning Algorithms: the Case of Ethio Telecom | en_US |
| dc.type | Thesis | en_US |