Mobile privacy is broadly concerning in the mobile big data era, as user mobility behaviors are privacy-sensitive and unique. User identification attacks consist of one of the most critical privacy concerns on mobile big data. In this paper, we study mobile privacy in terms of user identifiability from the perspective of privacy adversaries. User identification in two datasets from the same data source or two different data sources is generally formulated as a linear assignment problem (LAP), in which the cost matrix of users is generated by a single distance measure. However, user identification via one single distance measure may lead to a large portion of false matches, especially when only a few users coexist across these two datasets. In addition, the cubic computational complexity of LAP limits the scale of user identification analysis. In this paper, we propose a multi-feature ensemble matching framework to improve the user identification precision based on a majority voting rule, by integrating multiple distance measures. The computational complexity of the proposed ensemble matching algorithm is an order of magnitude less than that of the single-distance based approach, which results from solving an LAP on a highly sparse matrix rather than a dense matrix. Experiments demonstrate the superior performance of our proposed scalable ensemble matching framework with respect to matching precision as well as the vulnerability of mobile network subscribers' privacy.