Bolstering user authentication: a kernel-based fuzzy-clustering model for typing dynamics
Keywords:anomaly detection, data mining, fuzzy clustering, keystroke biometrics, kernel function, machine learning, static authentication
In most information systems today, static user authentication is accomplished when the user provides a credential (for example, user ID and the matching password). However, passwords appear to be the most insecure authentication method as they are vulnerable to attacks chiefly caused by poor password hygiene. We contend that an additional, non-intrusive level of security can be achieved by analyzing keystroke biometrics and coming up with a unique biometric template of a user's typing pattern. The paper proposes a new model for representing raw keystroke data collected when analyzing typing biometrics. The model is based on fuzzy sets and kernel functions. The corresponding algorithm is developed. In the static authentication problem, our model demonstrated relatively higher performance than some classic anomaly-detection algorithms, such as Mahalanobis, Manhattan, nearest neighbor, outlier counting, neural network, and the support-vector machine.