2016 International Conference on Artificial Intelligence and Computer Science (AICS 2016) ISBN: 978-1-60595-411-0 Study on Keystroke Dynamic with Feature of Pressure Qian-hong YAN 1 , Wei-kang WANG 2 , Ruo-xi QIN 3 , Hao-tian JIANG 4 , Bo-rui YANG 5 and Bao-chang ZHANG 6,* 1-6 BeiHang University, XueYuan Road No. 37, HaiDian District, BeiJing, China *Email: bczhang@buaa.edu.cn Keywords: Keystroke dynamic, Support vector machine, Hidden markov model. Abstract. This paper introduces a new database structure which is different from the existed one that only consists of press and release time. To improve, we add eigenvalue of pressure to it and we compare the results of new database structure with the old one to clarify whether the adding eigenvalue of pressure can improve the efficiency of the system. In addition, we come up a different keystroke dynamics method, Hidden Markov Models, to examine the efficiency of the new database structure and compare it with the results from another similar one, One-Class Support Vector Machine, to figure out whether Hidden Markov Models will behave better. Introduction The degree of informatization of modern society is rapidly deepening. Our life is fulfilled with all kinds of information. Therefore, the security problem should be paid more and more attention on. Nowadays pattern recognition technology has a plenty of applications and was applied in our daily lives. Among those methods, the most popular methods are Bio-metric authentication methods such as face recognition, fingerprint identification and palm print recognition [4]. However, all of them need complicated external devices. On the contrast, keystroke dynamics authentication is a simple way to authenticate without additional hardware [5,14]. Research on keystroke dynamics was started long time before, but it is still an on-going research. Traditional database of keystroke dynamics only contains press time and release time, which display the rhythm of the typing manner. However, many smart phones already have the ability to test the pressure of people when they are typing, so we add the new feature of the keystroke pressure on the basis of old features and hope to observe the improvement of the recognition rate. Hidden Markov Models are applied as a method of keystroke dynamics to compare the efficiency with One-Class Support Vector Machine to find out the better way to identify the characters on keyboard. Keystroke Dynamic Data Preparation Our experiment data consists of two parts: time and pressure. It is presented in the form of a sequence: P1,R1,P2,R2,…,Pm,Rm,T1,T2,…,Tm. Pi (i=1,2,…,m) represents the press time, which is defined as the time when user stroke the i th key. Ri represents the release time, which is defined as the time when user release the i th key [1]. Ti represents the pressure value when the i th key is pressed. From this set of raw data series, we can extract three types of features: flight time, dwelling time and the key pressure. So the time of keystroke duration is Ri-Pi (dwelling time), the time of keystroke interval is Pi+1-Ri (flight time) [6] and the key pressure is Ti. As a result, the sequence extracted from the raw data is: 475