Sim Box detection using machine learning techniques

Abstract

Subscriber Identity Module (SIM) box fraud causes severe International Direct Dialing (IDD) revenue losses to network operators and also causes harm to their brand names due to its low Quality of Service (QoS). Therefore, it is very important to prevent this fraud by disconnecting SIM numbers used for SIM box fraud with smaller call attempts to hinder fraudsters from generating revenue. However, detection of SIM box fraud numbers for disconnection consumes lots of work hours as it is necessary to minimize the impact on genuine subscribers. Call Detail Record (CDR) analysis is one of the methods to detect SIM numbers used in the SIM box. Machine learning techniques for fast data analysis are suitable for detecting SIM box numbers using CDR. This research was carried out to accurately identify suitable machine learning methods to detect SIM box fraud numbers. Subscriber profiles are created using CDR to train and evaluate machine learning models. Support vector classifier, decision tree classifier, and K-nearest neighbors classifier algorithms show better results from different types of machine learning models that were trained and evaluated to identify suitable algorithms to detect accurately. A weighted linear combination of these algorithms was used to get the best result with a macro F1-score value of 0.89 using the ensemble voting classifier algorithm. The result confirms that machine learning models can be used to detect SIM box fraud numbers accurately