基于约束集的半监督未知协议识别方法OA

Due to the absence of explicit prior features in unknown protocols and the scarcity of labeled samples,traditional identification approaches—such as port-based methods and deep packet inspection—as well as supervised learning techniques that rely heavily on large labeled datasets,become ineffective for traffic classification.To overcome these limitations,this paper proposes a semi-supervised unknown proto-col identification method based on constraint sets.The approach constructs multidimensional features by combining statistical features with protocol fingerprint features and applies constrained Laplace scores for feature selection.In addition,constraint information is derived from the correlation between labeled samples and real network traffic,and then incorporated into the K-means clustering process to enhance the recognition accuracy and efficiency.Experimental results demonstrate that this method significantly im-proves clustering performance compared with traditional clustering algorithms;in scenarios with multiple unknown protocols,the recognition accuracy exceeds 97%,providing a new solution for unknown protocol identification.