The paper describes an implementation of a behavioral authentication system, working on sparse geographical data generated by mobile devices in the form of CDR logs. While providing a review of state of the art w.r.t. sensors and measures that can be used when creating a system detecting anomalies in the user behavior, it also describes domain specific authorization methods focusing on the user mobility. The trajectory based stay-extraction model is utilized to build user mobility patterns, upon which the anomaly detection model measures the repeatability of human behavior in dimensions of: geography, time and sequentiality. The goal is to measure the extent to which the geographical aspect of the human mobility can be used in behavioral biometrics' systems i.e. in which scenarios geography may enable to describe (and differentiate between) user patterns - based on anomaly detection in cases resembling real life scenarios (phone theft or sharing between users). The research methods developed may be implemented on mobile devices to benefit from multiple sensors data in the authentication processes. The model is evaluated on a large telecom dataset, with the use of similarity classes, what allows measuring the accuracy of the model in real-life scenarios and provides benchmarking guidelines for the future work on the topic.