Subscription Fraud Prevention in Telecommunication using Deep Learning Approach: The Case of Ethio Telecom
| dc.contributor.advisor | Rosa, Tsegaye (PhD) | |
| dc.contributor.author | Gebremeskel, Aregay | |
| dc.date.accessioned | 2022-02-15T07:33:40Z | |
| dc.date.accessioned | 2023-11-04T15:13:05Z | |
| dc.date.available | 2022-02-15T07:33:40Z | |
| dc.date.available | 2023-11-04T15:13:05Z | |
| dc.date.issued | 2021-12 | |
| dc.description.abstract | With the rapid development of telecommunication services, telecommunication fraud is a challenging issue that fraudsters continuously participate in abusing the services to get undeserved benefit. For telecommunication operators who have no way of identifying mechanisms for their customers‟ legitimacy during service subscription, could lead to subscription fraud which is the cause for other fraud types. The purpose of the research is to prevent subscription fraud by identifying customer‟s face using face recognition techniques and propose the deployment scenario in telecommunication industries especially for those who have customer portraits. This study uses deep learning algorithms to build a face recognition model using a dataset of 124 identities collected from the Internet, personal image gallery and capturing from some friends. These identities have two different datasets; with 2596 images and 1240 images. The former dataset split into 80% and 20% ratio whereas the later one is vice versa provided that each identity in the training set has two images. Two separate experiments have been conducted for the first dataset; CNN with transfer learning, and MTCNN with FaceNet and SVM, whereas in the second dataset, the experiment with better accuracy in the first dataset is further retrained. Experimental results obtained an accuracy of 72.54% for CNN with transfer learning and 99.24% for MTCNN, FaceNet and SVM in the first dataset. The latter set of algorithms has achieved an accuracy of 98.69% in the second dataset which is recommended as a better solution. Finally, feasibility of the model deployment scenario has been analyzed and proposed by assessing the necessary requirements used for implementation. | en_US |
| dc.identifier.uri | http://etd.aau.edu.et/handle/123456789/30102 | |
| dc.language.iso | en_US | en_US |
| dc.publisher | Addis Ababa University | en_US |
| dc.subject | Telecommunication Fraud | en_US |
| dc.subject | Subscription Fraud | en_US |
| dc.subject | ethio telecom | en_US |
| dc.subject | Deep Learning | en_US |
| dc.subject | Face Recognition | en_US |
| dc.subject | CNN | en_US |
| dc.subject | Transfer Learning | en_US |
| dc.subject | MTCNN | en_US |
| dc.subject | FaceNet | en_US |
| dc.title | Subscription Fraud Prevention in Telecommunication using Deep Learning Approach: The Case of Ethio Telecom | en_US |
| dc.type | Thesis | en_US |