Adaptive Deep Feature Fusion for Continuous Authentication with Data Augmentation

Mobile devices are becoming increasingly popular and playing significant roles in our daily lives. Insufficient security weak protection mechanisms, however, cause serious privacy leakage of the unattended devices. To fully protect mobile device privacy, we propose ADFFDA, a novel continuous authentication system using an Adaptive Deep Feature Fusion scheme for effective feature representation, transformer-based GAN Data Augmentation, by leveraging smartphone built-in sensors accelerometer, gyroscope magnetometer. Given normalized sensor data, ADFFDA utilizes consisting generator CNN-based discriminator to augment training data CNN training. With augmented especially designed based on ghost module bottleneck, extracts deep features from three trained CNN, exploits adaptive-weighted concatenation method adaptively fuse CNN-extracted features. Based fused features, authenticates users one-class SVM (OC-SVM) classifier. We evaluate performance terms efficiency GAN, GAN-based augmentation, architecture, fusion, OC-SVM The experimental results show that obtains best w.r.t representative approaches, achieving mean equal error rate 0.01%.