Kolmogorov-Smirnov test for keystroke dynamics based user authentication

In this paper a novel user authentication method is proposed based on analyzing the keystroke patterns. Instead of using Nearest Neighbor classification with Dynamic Time Warping (DTW) distance measure, the typing dynamics are classified by the Kolmogorov-Smirnov test. First the typing pattern is translated into a sequence of holding times and latencies between consecutive keystrokes. In order to increase the classification performance, an adaptive preprocessing is used to get rid of the outliers. Then the Kolmogorov-Smirnov test is applied to classify the observed pattern. To further increase the correct classification ratio, semi-supervised self training methods are applied. One of the key objectives of the paper is to analyze the performance with respect to the length of typed characters, as in the case of authentication systems users cannot be asked to type long texts. The simulations have demonstrated that the proposed method performs better than the 1 Nearest Neighbor DTW classifier on all text lengths. It has also been shown that when the length of the typed text reduces to 10 characters, then the classification ratio sinks to 50% from 90.5% achieved in the case of longer texts in the range of 200 characters.