A Privacy-Preservation-Based Secured Model for Detection Network Crawlers

Crawler detection is crucial for Network Service Provider (NSP), and the federated learning-based approach to crawler detection has been under vigorous investigation. Nevertheless, training crawler detection models locally with NSP training server-side data is ineffective and expose privacy risks. To address this problem, we propose a privacy-preserving and communication-efficient model for decentralized federated learning by combining NSP training server-side and central server-side aggregation. Specifically, we developed an enhanced federated learning algorithms to train the model, avoiding data transfer burdens. We design perturbed local model parameters and corresponding method to mitigate privacy risks during upload. We developed an innovative data processing algorithm that effectively enables the server to obtain structured data with spatiotemporal information by analyzing user access time and resources. Finally, we improved the time series detection model by fusing the output of each time step with the hidden layer, effectively improve the robustness of the detection model. Evaluation on two datasets under two network crawler scenarios shows substantial improvement of accuracy while ensuring privacy protection.