A DEEP LEARNING FRAMEWORK WITH ATTENTION AND BIG-STEP CONVOLUTION FOR NETWORK TRAFFIC ANOMALY DETECTION

Abstract

Finding unusual traffic is essential for both network security and service quality. A big-step convolutional neural network traffic detection model based on the attention mechanism is developed since the single dimension of the detection model and feature similarity make it very difficult to identify anomalous traffic. First, the raw traffic is preprocessed and mapped into a two-dimensional greyscale picture after the network traffic characteristics are examined. After that, histogram equalisation is used to create multi-channel greyscale pictures, and an attention mechanism is added to give traffic characteristics varying weights in order to improve local features. Lastly, traffic characteristics of various depths are extracted by combining pooling-free convolutional neural networks, which improves convolutional neural network flaws including overfitting and local feature omission. A balanced public data set and an actual data set were used for the simulation experiment. The suggested model is contrasted with ANN, CNN, RF, Bayes, and two more recent models, using the widely used method SVM as a baseline. In experiments, 99.5% accuracy is achieved using several classifications. The best anomaly detection is found in the suggested model. Additionally, the suggested approach beats other models in F1, recall, and accuracy. It is shown that the model is strong and resilient to many complicated situations in addition to being effective at detection.