Because mobile devices are easily lost or stolen, continuous authentication is extremely desirable for them. Behavioral biometrics provides non-intrusive continuous authentication that has much less impact on usability than active authentication. However single-modality behavioral biometrics has proven less accurate than standard active authentication. This thesis presents a behavioral biometric system that uses multi-modal fusion with user data from touch, keyboard, and orientation sensors. Testing of ve users shows that fusion of modalities provides more accurate authentication than each individual modalities by itself. Using the BayesNet classification algorithm, fusion achieves False Acceptance Rate (FAR) and False Rejection Rate (FRR) values of 9.65% and 2% respectively, each of which is 8% lower than the closest individual modality.
展开▼
