Feature ranking for effective continuous user authentication using keystroke and mouse dynamics with the cat recurrent neural model
Princy Ann Thomas, Preetha Mathew Keerikkattil · International Journal of Biometrics · 2025
Behavioural biometric modalities such as keystroke and mouse dynamics are ideal for continuous user authentication due to their non-intrusive quality. The success of the authentication framework is largely determined by the discriminative power of the features used. It is critical to be able to select the necessary discriminative features for optimal authentication performance. In this research, we implement multiple ranking algorithms on features derived from temporal information of keystroke and mouse dynamics to distinguish their discriminative capacity. The ranked features are then employed for continuous authentication using the cat recurrent neural model (CRNM) to optimise the search space and authenticate users. The experimental results given in this work propose a strategy for developing commercially deployable continuous authentication systems with broad applicability. Experiments are carried out with filter, wrapper, and embedded feature ranking approaches, and authentication outcomes are compared with the CRNM framework. The findings indicate that discrimination is manifested in uncommon rather than normal user conduct. Furthermore, it is discovered that applying feature ranking reduces authentication time from 198 seconds to 138 seconds and improves accuracy from 98.25% to 99.21%.