Mobile User Identification across Domains Based on the Stacking Method in Ensemble Learning

Abstract

Given the behavioral data of a mobile device, and finding similar devices among massive candidates, this is actually a user identification problem using mobile device data. This problem is essential in many practical applications, such as authorized user identification, personalized recommendation and privacy protection. Most of the previous work in this problem only used the data from a single domain, such as trajectory data or app data, and most of their experimental datasets are in an ideal environment with a small number of users, high sampling rate and high accuracy. However, it is often difficult to obtain this kind of data in real-world activities. To address these limitations, we introduce a new framework for mobile user identification, divided into a recall module and a sorting module. In the sorting module, we design and combine multiple similarity comparison algorithms based on stacking method in ensemble learning. We evaluate the framework performance on a realworld anonymous dataset from numerous mobile devices, containing the behavioral data across several domains but with a low sampling rate and accuracy. Our framework achieves an AUC of 0.917 and F <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</sub> score of 0.839 which outperforms the state-of-the-art methods.