NETWORK TRAFFIC ANOMALY DETECTION BASED ON GLOBAL HIERARCHICAL FEATURE FUSION AND MULTI-TASK LEARNING

Aimed at the problem of weak representation and generalization ability of current deep learning-based methods, a method for detecting anomaly network traffic based on global hierarchical feature fusion and multi-task learning is proposed. We segmented the network traffic in units of sessions and fused the spatial and temporal features of session streams extracted by global hierarchical feature fusion framework parallelly. A multi-task learning framework was designed in which the multi-classification of conversation records was the main task, and the multi-classification of conversation flow and whether the conversation flow pair was contextual are auxiliary tasks. We inputted session stream pair for training and prediction. Experimental results on TON-IOT dataset show that accuracy rates of binary classification and multi-classification are 94.35% and 91.96%. Compared with the comparison method, it maintains the lowest false alarm rate when accuracy and precision are optimal.