Improved Keystroke Authentication Accuracy Based On Statistics And Weight

Nowadays, people need to prove their identity in many cases, especially in the electronic security transactions, the identity authentication is particularly important. However, the existing authentication methods, like the password, USB Key, smart cards, dynamic token, etc, all have their defects. On one hand, password authentication is vulnerable to password guessing and dictionary attacks, also the password is easy to be peeped by attacker. On the other hand, the additional hardware support is essential for the hardware authentication methods, which are not convenient to carry on, and result to the reduction of the utilization. All in all, the biometric authentication is reflected increasingly important. The purpose of the thesis is that to figure out an algorithm, which can meet the commercial application of keystroke authentication, that is to say that it has low FRR (False Rejection Rate, False Rejection Rate), FAR (False Accept Rate error through rate) and EER (Equal Error Rate, error rate). Keystroke authentication method is one of behavioral characteristics authentication, which overcomes the above drawbacks. Keystroke authentication is based on people’s keystroke characteristics, including the keystroke interval, keystroke duration, when the user is typing on the keyboard, and according to certain algorithms, such as neural networks, fuzzy math, to generate templates for the user’s characteristics, and then the keystroke authentication can enhance the authentication when the user logs in a system.Mainly two issues in the field of keystroke authentication algorithm are discussed in the thesis. Firstly, using a third-party data source from Carnegie Mellon University to provide the user the keystroke data information, then identifie multiple statistical keystrokes authentication algorithms,including Manhattan, Manhattan(filtered), Manhattan(scaled), Mahalanobis, Mahalanobis(Normed), Diff_Subspace, seq_dist, Statistics, Statistics_ex, Two-dimensional space algorithm and Combinations of probabilistic algorithm. At last, the experiment result indicates that the Manhattan (scaled) algorithm is better than other algorithms, its FAR=8.8%and FRR=13.0%. Secondly, in order to improve the accuracy of the keystroke authentication, a more efficient keystroke authentication algorithm is proposed in the thesis, and FAR=7.9%, FRR=6.2%.