In-app activity recognition from Wi-Fi encrypted traffic

In today’s world mobile applications have been widely used, which bring great convenience to people’s lives. However, at the same time user privacy is potentially threatened. This paper shows that a passive eavesdropper can identify fine grained user activities (known as in-app activities) by analysing encrypted traffic collected by sniffing a wireless network. Even though encryption protocols are used to secure communications over the Internet, side channel data such as frame length, inter arrival time and direction are still leaked from encrypted traffic. To identify in-app activities from this side channel data machine learning techniques are used. Furthermore, we show that just by observing only a small subset of encrypted traffic (rather than observing the entire transaction), one can identify in-app activities accurately. The proposed solution was evaluated with 51 in-app activities from three popular social networking apps and obtained high detection accuracy, 95.4% when Bayes Net algorithm is used.