A Malicious Traffic Detection and Classification Method using Multi-task Spatiotemporal Convolutional Networks for 5G Networks

With the development of 5G networks, malicious traffic poses an increasingly serious threat to network security. Because the overfitting problem significantly impacts single-task traffic detection, improving the detection model's correctness and generalization is essential. To this end, this paper introduces a multi-task learning framework integrating the Adaptive Neuro-Fuzzy Inference System (ANFIS) and proposes a Fuzzy Logic Enhanced Multi-Task Spatiotemporal Convolutional Network (FL-MTSTCN) to enhance malicious traffic detection. At first, we convert raw network data into one-dimensional arrays and then input them into a shared cell model to extract temporal and spatial characteristics of network traffic. Next, the model utilises the ANFIS module to improve its ability to handle uncertainty and enhance the model's robustness and decision-making capabilities in malicious traffic detection. Experimental results show that FL-MTSTCN outperforms other existing methods on most tasks with an average accuracy of more than 98% when performing multiple tasks simultaneously. Eventually, we deployed the FL-MTSTCN model on our designed 5G heterogeneous network management platform to achieve real-time detection and classification of 5G network traffic, which further verifies the effectiveness and practical value of the proposed FL-MTSTCN in practical applications.