Fig. 2From: Evaluating smartphone-based dynamic security questions for fallback authentication: a field studyTraining and test scheme for evaluation of the Bayes classifier. For training, we build 2 models: (1) a model for each legitimate user, (2) a community model for attackers excluding the corresponding legitimate user’s attacker data. Since it is unrealistic to assume that attackers’ data is known in advance by the system, we use this scheme for more realistic evaluation of the classifierBack to article page