Near-Real Time SIM-box Fraud Detection Using Machine Learning in the case of ethio telecom

The advancement of telecommunication era is rapidly growing, however, telecom fraudsters encouraged by the emerging of these new technologies. Interconnect bypass fraud is one of the most sever threats to telecom operators. Subscriber Identity Module Box (SIM-box) fraud is one of an interconnect bypass telecom fraud type and uses Voice over IP (VoIP) technology. In addition, it's difficult to detect such fraud types with Test Call Generation (TCG) and a traditional types of Fraud Management System (FMS). Both TCG and FMS easily bypassed by the fraudsters, telecom companies impacted by losing billions of dollars. In this study, Sliding Window (SW) aggregation mode is applied to provide a relevant dataset instance and reduce detection delay to one hour by using supervised Machine Learning (ML) algorithm. Three supervised ML classifier algorithms were used, namely Random Forest (RF), Artificial Neural Network (ANN), and Support Vector Machine (SVM) with the two validation techniques 10 -fold cross-validation and supplied test. Call Detail Record (CDR) data were collected, relevant attributes were selected and preprocessing such as data cleaning, integrating and aggregating tasks were performed. The experimental results depict that RF classifier using cross-validation on SW aggregation mode achieves a better classification accuracy (96 .2 %). ANN is placed on second with its overall performance accuracy and its detection delay, SVM algorithm using cross-validation exceeds the desired detection delay (49 ,965 second) with poor performance accuracy. RF classifier algorithm using SW aggregation mode overcomes the trade-off detection accuracy and detection delay.