I. Introduction
Various tools, such as security/quality-of-service enforcement devices and network monitors, rely on the knowledge of the application generating the traffic and thus are limited (or impaired) when this requirement is not completely satisfied. The process of associating network traffic with specific applications is known as Traffic Classification (TC) and has a long-established application in several fields [1]. Notwithstanding, TC is challenged by the massive diffusion of handheld devices (as supported by recent evaluations in Internet usage [2]), which is revolutionizing the nature and the composition of traffic traversing home and enterprise networks and connecting contents and services over the Internet. Thus, the necessity and the difficulty of mobile TC have both become consistently high nowadays, fueled (other than common drivers for TC) by valuable profiling information (e.g., to advertisers, insurance companies, and security agencies) [3], [4], while also implying privacy downsides (e.g., recognition of context-sensitive apps, such as health and dating ones, and in case of bring-your-own-device policies from companies).