Résumé:
The SIMBox is a telecommunications device that reroutes international voice calls
through the internet to a local SIM card within the device, making them appear as
local calls and bypassing standard network gateways. This Master’s thesis presents
a SIMBox Fraud Detector, an AI-powered web application for SIMBox fraud detec-
tion in mobile telecommunications networks. A thorough experiments with machine
learning techniques using the dataset CDR (Call Detail Records) from Djezzy mo-
bile operator show that the XGBoost model with Tomek+RUS undersampling exhibits
promising advantages in both detection and balance performance. XGBoost not only
achieved a precision and accuracy of 96% for fraudulent detection, but it also pro-
vided the best tradeoff between false positive rate and true positive rate.
