A new approach to securing passwords using a probabilistic neural network based on biometric keystroke dynamics

Abstract

Passwords are a common means of identifying an individual user on a computer system. However, they are only as secure as the computer user is vigilant in keeping them confidential. This thesis presents new methods for the strengthening of password security by employing the biometric feature of keystroke dynamics. Keystroke dynamics refers to the unique rhythm generated when keys are pressed as a person types on a computer keyboard. The aim is to make the positive identification of a computer user more robust by analysing the way in which a password is typed and not just the content of what is typed. Two new methods for implementing a keystroke dynamic system utilising neural networks are presented. The probabilistic neural network is shown to perform well and be more suited to the application than traditional backpropagation method. An improvement of 6% in the false acceptance and false rejection errors is observed along with a significant decrease in training time. A novel time sequenced method using a cascade forward neural network is demonstrated. This is a totally new approach to the subject of keystroke dynamics and is shown to be a very promising method The problems encountered in the acquisition of keystroke dynamics which, are often ignored in other research in this area, are explored, including timing considerations and keyboard handling. The features inherent in keystroke data are explored and a statistical technique for dealing with the problem of outlier datum is implemented.