Biometric is used as a main security fence in a computer system. Human’s biometrics can be categorized into three types: morphological, biological and behavioral. This study focuses on behavioral biometric by studying keystroke dynamics. Keystroke dynamics is used as the user’s second level verification for the systems that use the keyboard to login into a system. Soft biometric is a method by which we can identify human identity based on the physical characteristics or behaviors of a person that are naturally produced. A common problem of system intrusions is that the system fails to identify the real user who signs in using the keyboard when the login is correct. A user’s login information can be stolen, guessed, and recorded by special software such as a keylogger. There is a possibility that someone else tries to break into the system by using brute-force and dictionary attacks. To ensure and improve user’s verification who use the keyboard to enter their logins into the system, this study has used multi model fusion methods to combine keystroke dynamics and soft biometric. This study introduces new soft biometric elements in keystroke dynamics that is gender, race, region and educational level. Keystroke dynamics is used to distinguish typing patterns in each of these categories and uses these soft biometric features to further enhance the verification capabilities. Mathematical formulas such as mean, standard deviation and Scale Manhattan Distance have been used to obtain the score of the imposter and genuine user. Multi model fusion methods have been introduced by integrating the score values of several soft biometric elements from classification with scores obtained from imposter and genuine user to improve the verification process. For identification purposes, Support Vector Machine classification method is used to perform this classification for soft biometric identification. The results show that there are significant differences in the typing pattern in the region category. The highest accuracy achieved is 91.17% in this classification process for the region category. Equal error rate (EER) obtained from a multi model fusion approach is 11.33%. which is to incorporate four features of soft biometrics (gender, race, region and educational level) into the verification process. Overall, it is found that the multi fusion process introduced has managed to reduce the EER values for the purpose of verification in keystroke dynamics.
