Privacy-Preserving Sketch-Based Big Data Traffic Detection
Chuan Zhang, Yujing Liu, Xuhao Ren, Jiayi Xu, Yajie Wang, Liehuang Zhu · 2025
With the rapid growth of big data, network traffic detection has become a core task for network management and security. However, traditional methods face two significant challenges in large-scale data traffic processing: first, the statistics of tail traffic are inaccurate, which leads to the neglect of key traffic features, thereby weakening the effectiveness of anomaly detection; second, user traffic data faces the risk of privacy leakage. To address these problems, we propose a privacy-preserving sketch-based big data traffic detection scheme, named PPTD. By introducing Paillier homomorphic encryption technology and an improved sketch structure, the scheme processes large-scale traffic data and accurately evaluates the characteristics of tail traffic. Specifically, we achieve hierarchical modeling of low-frequency and high-frequency traffic through the collaboration of the random admission (RA) structure and two count-mean-min (CMM) Sketches, improving the statistical accuracy of tail traffic characteristics. Furthermore, the introduction of Paillier homomorphic encryption technology allows sensitive user data to be directly calculated in encrypted form. Security analysis demonstrates that PPTD effectively protects the privacy of user traffic data. Experiments show that, compared with existing methods, PPTD improves the accuracy of tail traffic detection in tests on real and simulated datasets.