Single Classifier And Its Applications, Keystroke Authentication System

One class classification is a machine learning approach different from the traditional pattern recognition approach where two or more class samples are required. However in some real-life cases, we can hardly, even not, get the samples of some classes, or have to pay costly price to obtain the so-needed samples, such as in the case of machinery malfunction. And while in other cases, the sizes of samples among classes are imbalance, such as medical diagnosis. All these problems are suitable for one class classification approach. In this article, after analyzing three typical approaches of one-class classifiers, we focus much on a generalization of Campbell and Bennett's linear-programming-based one-class classification approach to novelty detection, and extend it to a weighted formulation, then perform some experiments on the real datasets. The results show that the new approach gets better performance compared to the original counterpart. At last, we integrate the one class classification approach to the Windows NT logon authentication system to realize a practically useful keystroke authentication system with high performance.