MBBFAuth: Multimodal Behavioral Biometrics Fusion for Continuous Authentication on Non-Portable Devices

Abstract

Continuous authentication based on behavioral biometrics is effective and crucial as user behaviors are not easily copied. However, relying solely on one behavioral biometric limits the accuracy of continuous authentication. Therefore, a continuous authentication system based on multimodal behavioral biometrics fusion is proposed in this study, which fuses three modalities: contextual behavior, mouse behavior, and information interaction behavior. The multimodal dataset of user behavior is collected through a self-built website, and the behavioral feature sets for each modality are then created. An improved generative adversarial network method is used to align the datasets of the three modalities. The autoencoder with long short-term memory is employed for unsupervised anomaly detection of time-series behaviors and enables continuous authentication for each modality. The multimodal fusion is achieved using the meta-model of the stacked generalization method, and the final decision for continuous authentication is then determined. The experimental results demonstrate that the proposed multimodal fusion method significantly outperforms the unimodal and provides an effective way to improve the accuracy and user-friendliness of continuous authentication. This study offers insights into user behavior analysis, behavioral anomaly detection, and multimodal behavior fusion.