SMS-Watchdog: Profiling Social Behaviors of SMS Users for Anomaly Detection

Abstract

With more than one trillion mobile messages delivered worldwide every year, SMS has been a lucrative playground for various attacks and frauds such as spamming, phishing and spoofing. These SMS-based attacks pose serious security threats to both mobile users and cellular network operators, such as information stealing, overcharging, battery exhaustion, and network congestion. Against the backdrop that approaches to protecting SMS security are lagging behind, we propose a lightweight scheme called SMS-Watchdog that can detect anomalous SMS behaviors with high accuracy. Our key contributions are summarized as follows: (1) After analyzing an SMS trace collected within a five-month period, we conclude that for the majority of SMS users, there are window-based regularities regarding whom she sends messages to and how frequently she sends messages to each recipient. (2) With these regularities, we accordingly propose four detection schemes that build normal social behavior profiles for each SMS user and then use them to detect SMS anomalies in an online and streaming fashion. Each of these schemes stores only a few states (typically, at most 12 states) in memory for each SMS user, thereby imposing very low overhead for online anomaly detection. (3) We evaluate these four schemes and also two hybrid approaches with realistic SMS traces. The results show that the hybrid approaches can detect more than 92% of SMS-based attacks with false alarm rate 8.5%, or about two thirds of the attacks without any false alarm, depending on their parameter settings.