International Revenue Sharing Fraud (IRSF) Detection Using Data Mining Techniques: The Case of Ethio Telecom
| dc.contributor.advisor | Ephrem, tesh | |
| dc.contributor.author | Alamarew, Digafe | |
| dc.date.accessioned | 2022-02-10T10:47:52Z | |
| dc.date.accessioned | 2023-11-04T15:13:03Z | |
| dc.date.available | 2022-02-10T10:47:52Z | |
| dc.date.available | 2023-11-04T15:13:03Z | |
| dc.date.issued | 2022-01 | |
| dc.description.abstract | One of the most often expressed concerns in the telecom industry is international revenue sharing fraud. Fraudsters are motivated to generate traffic for their services without paying the originating network. Existing detection methods include monitoring call patterns and blocking high-risk range lists obtained from various telecom associations. However, these approaches have issues that make them ineffective, time-consuming techniques, leading to financial losses before the call is blocked as they are frequently bypassed, renew high-risk range list numbers repeatedly, and make lists out of date. The goal of this paper is to develop an international revenue sharing fraud model for real-time detection of missed call fraud generation schemes using an international voice call detail record in an hourly manner to minimize detection time as well as revenue loss. To do so, data is collected, data preprocessing is performed, and relevant attributes are determined. As classifiers, support vector machines, artificial neural networks, and random forests are utilized to classify the data set for fraud and normal call transactions. The findings indicate that random forest techniques outperform in terms of fmeasure, accuracy, receiver operating characteristic curve, the time required for building, and inference time in both training modes (percentage split and 10- cross-validation). It performed with 97.00% accuracy and also achieved comparable accuracy to the state of the art. Consequently, the support vector machine and neural network multilayer perceptron classification algorithms are found to be the next best after the random forest in terms of overall performance metrics. However, the support vector machine classifier in both test modes exceeds the acceptable detection delay in the classification process. | en_US |
| dc.identifier.uri | http://etd.aau.edu.et/handle/123456789/29994 | |
| dc.language.iso | en_US | en_US |
| dc.publisher | Addis Ababa University | en_US |
| dc.subject | International revenue sharing fraud | en_US |
| dc.subject | International premium rate number | en_US |
| dc.subject | Classification | en_US |
| dc.subject | Telecom fraud | en_US |
| dc.subject | Fraud detection | en_US |
| dc.subject | Missed calls | en_US |
| dc.title | International Revenue Sharing Fraud (IRSF) Detection Using Data Mining Techniques: The Case of Ethio Telecom | en_US |
| dc.type | Thesis | en_US |