One-class classifiers : a review and analysis of suitability in the context of mobile-masquerader detection

摘要: One-class classifiers employing for training only the data from one class are justified when other classes is difficult to obtain. In particular, their use in mobile-masquerader detection, where user characteristics classified as belonging legitimate or impostor class, and collecting originated impostors problematic. This paper systematically reviews various one-class classification methods, analyses suitability context of detection. For each method, its sensitivity errors set, computational requirements, considered. After that, category features used masquerader suitable identified.