A continuous smartphone authentication method based on gait patterns and keystroke dynamics

• Published in: Journal of ambient intelligence and humanized computing Vol. 10; no. 11; pp. 4417 - 4430
• Main Authors: Lamiche, Imane, Bin, Guo, Jing, Yao, Yu, Zhiwen, Hadid, Abdenour
• Format: Journal Article
• Language: English
• Published: Berlin/Heidelberg Springer Berlin Heidelberg 01.11.2019; Springer Nature B.V
• Subjects: Accelerometers; Accuracy; Algorithms; Artificial Intelligence; Authentication; Behavior; Biometrics; Cellular telephones; Classifiers; Computational Intelligence; Datasets; Energy consumption; Engineering; Gait; Interactive computer systems; Machine learning; Methods; Multilayer perceptrons; Original Research; Personal identification numbers; Physiology; Rejection rate; Robotics and Automation; Robustness (mathematics); Security; Sensors; Smartphones; User Interfaces and Human Computer Interaction; Walking; Behavioral biometrics; Keystroke dynamics; Gait analysis; Continuous authentication; Smartphone sensing
• ISSN: 1868-5137; 1868-5145
• DOI: 10.1007/s12652-018-1123-6

Behavioral biometrics, such as gait patterns and keystroke dynamics, have been becoming increasingly used in human identity recognition research for enhancing the smartphone security. A new multimodal authentication method able to strengthen the smartphone authentication system is proposed in this paper. The proposed mechanism acquires gait patterns from the accelerometer, as well as keystroke dynamics, continuously without user intervention through simultaneous walk and text input. More specifically, features are extracted from both modalities. Afterward, a feature level fusion method is applied to build a multimodal biometrics profile for the user. Fused feature vectors are subjected to the sequential floating forward selection algorithm to reduce their dimensions as well as the computational complexity. The effectiveness of the proposed method is examined through a real multimodal dataset collected from 20 subjects under various scenarios, using different machine learning classifiers. The experimental results achieved a promising accuracy of 99.11% when using multilayer perceptron classifier with the average false acceptance rate, false rejection rate and equal error rate values of 0.684%, 7%, and 1%, respectively. Furthermore, the security strength of the proposed method was evaluated against two types of attacks, the zero-effort attack and minimal-effort mimicking attack. Results demonstrate that our approach represents a robust and secure authentication solution.