Ephrem, Teshale (PhD)Samuel, Mekasa2022-02-152023-11-042022-02-152023-11-042022-01http://etd.aau.edu.et/handle/123456789/30092Mobile roaming data-internet fraud, committed on visitor networks is a continued challenge and significant source of revenue losses for telecommunications societies including customers. The actually introduced prevention and detection mechanism have limitations in protection of the service. In this study, we used different data-sets and build roaming mobile data fraud detection model. Three supervised machine learning algorithms: Artificial Neural Network (ANN), Support Vector Machine (SVM) and J48 decision tree (J48 DT) where used to build model from each data-set. The model performance was computed based on different metrics. The model with merged data-set (roaming in and roaming out) achieved better performance and J48 DT is resulted greater in accuracy of 99.50, average F1_Score 99.00 and ROC 99.30. For compiled usage behavior exceeds the detection of such fraud, organization better to periodically analysis of data rather than waiting for TAP file-user usage from visited network in addition to revising roaming agreement.en-USUser behaviorMobile data roaming fraud detectionMobile data usageMachine learning algorithmsMachine learning toolsHome networkVisited networkThesis