加密流量分析中的特征建模与学习范式研究综述OA

The proliferation of encrypted traffic,driven by advanced encryption protocols and growing privacy aware-ness,renders traditional plaintext-based analysis methods obsolete.Accurately identifying traffic types and malicious behaviors in this encrypted context has become a critical challenge for network management and security monitoring.However,current approaches are hindered by two key bottlenecks:insufficient feature representation and high data annota-tion costs.To address this challenge,this survey systematically reviews the impact of mainstream encryption protocols on feature observability and establishes a general framework for classification and detection tasks.It further clarifies the con-cept of feature modeling,focusing on key stages such as feature extraction,selection,and multi-feature fusion.Under diminishing supervisory signals,learning paradigms are defined as training strategies with different levels of supervision,covering supervised,semi-supervised,and self-supervised learning for traffic classification and detection.Finally,the dis-cussion extends to key practical challenges,including model lightweighting,domain adaptation,and adversarial robust-ness,concluding with an outlook on future research directions.