A Cascaded Broad Learning Network Embedded Image Features for Malware Traffic Classification

Malware traffic classification (MTC) is a critical technology in network security, Internet of Things (IoT) devices, and big data traffic, and has been mainly used for encrypted malicious traffic detection and various network traffic classification. The traditional MTC technology mainly relies on payload-based deep packet inspection methods and port-based identification detection techniques. Meanwhile, In recent years, many studies have begun to explore the application of machine learning (ML) and deep learning (DL) in MTC tasks. However, ML methods require extracting excellent feature information, which can be challenging for traffic detection tasks. DL methods require excellent computing power equipment, such as GPUs, but also have the drawbacks of long training time and poor interpretability. Consequently, this paper proposes an image feature embedding cascaded broad learning network (IFCBLN) to address the shortcomings of ML and DL in MTC tasks. The proposed IFCBLN method utilizes multiple cascaded broad learning networks (BLN) combined with image feature extractors to improve the original BLN. Meanwhile, it achieves good performance in MTC tasks. In the datasets of USTC-TFC2016 and ISCXVPN2016, the proposed IFCBLN achieves better performance than ML and DL methods. The best results of the algorithm achieve an accuracy of 99.543% and 90.93%, respectively. The proposed method reduces computational overhead by 98% compared to the DL method.