Modified Distance Metric That Generates Better Performance For The Authentication Algorithm Based On Free-Text Keystroke Dynamics

The authentication process can be categorized by the number of incorporated factors: something you know, like a username and a password, something you have, like card, token or something you are, like biometrics. Keystroke dynamics has been pointed out as a practical behavioral biometric feature that does not require any additional device for scale up user authentication. The input data of an authentication system based on keystroke dynamics are the typing times on the keyboard. Given that typing times result in time vectors, and these must be compared to see the similarities between them to validate the user. Algorithms of dynamic authentication can be divided into three major groups: estimation of metric distances, statistical methods and machine learning. The paper aims to analyze the possibilities of increasing the efficiency of an authentication algorithm based on keystroke dynamics, in the sense of reducing the value of the Equal Error Rate (EER). The distance method is used to calculate the similarity between users. The paper (1) analyzes the optimal number of di-graphs, (2) analyzes the optimal time combinations generated by a di-graph to be used and, finally, (3) analyzes the possibility of modify the distance calculation metric. These analyzes aim to reduce the error rate generated by an authentication system based on free-text keystroke dynamics. The authors propose a modification of the Manhattan distance calculation formula that generates better performances, EER was improved by 38.53%. The EER value obtained from the modified metrics is 3.27%, compared to 5.32% obtained with the classic Manhattan formula.