Proximity User Identification Using Correlogram

Overview & Synopsis

This paper represents a technique, applying user action patterns in order to distinguish between users and identify them. In this method, users' actions sequences are mapped to numerical sequences and each user's profile is generated using autocorrelation values. Next, cross-correlation is used to compare user profiles with a test data. To evaluate our proposed method, a dataset known as Greenberg's dataset is used. The presented approach is succeeded to detect the correct user with as high as 82.3% accuracy over a set of 52 users. In comparison to the existing methods based on Hidden Markov Model or Neural Networks, our method needs less computation time and space. In addition, it has the ability of getting updated iteratively which is a main factor to facilitate transferability.