ST-Mamba: Spatio-Temporal Feature-Based Encrypted Traffic Analysis Using Mamba Network

With the widespread adoption of encrypted traffic in network communications and the Internet of Things (IoT), security analysis and monitoring of network traffic face new challenges. The presence of resource-constrained devices and complex traffic patterns reveals the limitations of existing classification methods in computational efficiency, representation, and generalization. To address these issues, we propose an efficient encrypted traffic classification model named ST-Mamba. In the preprocessing stage, we enhance the representation capability of encrypted traffic by integrating its spatio-temporal features and eliminating irrelevant biases. In the classification stage, we design a novel lightweight module based on the Mamba architecture to efficiently learn general traffic representations and capture essential features. This design reduces computational complexity from quadratic to linear, significantly improving processing efficiency. Finally, through evaluations of five classification tasks on five publicly available datasets—ISCX-VPN, USTC-TFC, ISCX-Tor, CICIoT 2022 and CSTNET-TLS 1.3—the results demonstrate that the ST-Mamba model achieves recognition accuracy exceeding 97.8% in all tasks. Moreover, compared to other existing models, ST-Mamba strikes a better balance between memory optimization and computational efficiency.