Truncated Lanczos-TSVD: An Effective Dimensionality Reduction Algorithm for Detecting DDoS Attacks in Large-Scale Networks

Sizhang Li, Jing Xu, Peng Liu, Xue Li, Puming Wang, Xin Jin, Shaowen Yao

Abstract: With the rapid development of Big Data, the scale and complexity of network data have significantly increased. Consequently, detecting DDoS attacks has become increasingly difficult. In this context, traditional machine learning has a limited ability to detect DDoS attacks, resulting in lower detection rates and efficiency. Therefore, there is an urgent need to address the problem of detecting DDoS attacks in large-scale network data while reducing computational costs and memory usage. To address this issue, the study adopted the following strategies: (I) Representing large-scale network data with tensors; (II) Applying the Truncated Lanczos-TensorSVD (TLanczos-TSVD) algorithm to reduce dimensions and remove noise from high-dimensional data; (III) Developing a DDoS attack detection framework that combines (I), (II), and the XGBoost classification model. To evaluate the framework's performance in detecting DDoS attacks and the efficiency of the denoising algorithm, multiple comparative experiments were conducted. These results indicate that the framework achieved an accuracy rate of 99.15%, which is the highest among all tested methods. Furthermore, it managed to maintain low costs and minimal memory usage. In addition, the framework demonstrated excellent detection performance on datasets of varying sizes, highlighting its strong robustness. In conclusion, this study proposed an efficient DDoS attack detection framework.