earticle

영어

Web security middle ware system has two functions, which are user authentication and security access control. User authentication is to verify a valid user. User authentication is commonly based on the method of password. The losing, opening, and hacking password causes a critical security problem. This research is focused on developing a neural network approach for user authentication using keystroke. Two-phase approach for user authentication and verification is designed. This first-phase is based on pattern analysis of keystroke of user ID and password. The second-phase also employs patterns analysis on keystroke of randomly generated characters complementarily. Keystroke includes the information on duration of pressing a character and delay interval time between pressing a character and next character. The proposed approach basically relied on the individual keystroke pattern. Conventional back-propagation learning algorithm is adopted with multi-modal sigmoid function. Various experimental analysis are performed to verify the presented method. Seven characters are used as keystroke pattern data. Classical minimum distance classifier and single perceptron are compared to back-propagation. Back-propagation algorithm has superior performance among the three approaches having 0.02441 FRR and 0.0288% FAR, which are close to other recognition rate. After user authentication is performed, Access control is designed for controlling not permitted or not allowed access. The presented neural network based user authentication is applied for Web security middle ware system.