A Continuous Authentication Approach for Mobile Crowdsourcing based on Federated Learning

With the widespread use of smartphones and wearable devices, Mobile Crowdsourcing (MCS) has become a powerful method for gathering and processing data from various users. MCS offers several advantages, including improved mobility, scalability, cost-effectiveness, and the utilization of collective human intelligence. However, ensuring the authenticity of users throughout the data collection process remains a challenge. Current authentication methods, such as traditional PIN codes, two-factor authentication, and biometric authentication, often struggle to provide continuous verification while adequately protecting user privacy. This paper addresses this issue by proposing a new continuous authentication approach based on Federated Learning. This approach combines continuous identity verification with privacy preservation benefits, allowing for the ongoing validation of user authenticity during data collection while improving authentication accuracy. We also discuss the non-Independently and Identically Distributed issue in Federated Learning and employ transfer learning techniques based on feature extraction to enhance the performance of the authentication models. Extensive experiments using various datasets were conducted to evaluate the effectiveness of our proposed method. The results of this study demonstrate our potential to enhance the security and privacy of MCS systems.