Application of keystroke analysis to mobile text messaging

The ability to verify the identity of a mobile subscriber will become increasingly important as the services and information that can be accessed by a cellular handset becomes more wide-ranging and sensitive. Current user authentication for mobile handsets is provided by the Personal Identification Number (PIN), which has a number of inherent weaknesses, such as users writing them down, telling other people and the option of not using it in the first place. As such non-intrusive, continuous and stronger authentication mechanisms are required. This paper presents the second phase of an investigation into the feasibility of one such technique, the use of keystroke analysis. Specifically, this paper evaluates the potential to authenticate subscribers by the way in which they type text messages. The study trials a number of classification algorithms based upon feed-forward multiple layer perceptron neural networks, evaluating their ability to differentiate between authorised and impostor users. Promising results have been observed, with average classification results of 18% Equal Error Rate (EER) and individual users achieving an EER as low as 3.2%.