Abnormal Traffic Detection Based on Attention and Big Step Convolution

Abnormal traffic detection is critical to network security and quality of service. However, the similarity of features and the single dimension of the detection model cause great difficulties for abnormal traffic detection, and thus a big-step convolutional neural network traffic detection model based on the attention mechanism is proposed. Firstly, the network traffic characteristics are analyzed and the raw traffic is preprocessed and mapped into a two-dimensional grayscale image. Then, multi-channel grayscale images are generated by histogram equalization, and an attention mechanism is introduced to assign different weights to traffic features to enhance local features. Finally, pooling-free convolutional neural networks are combined to extract traffic features of different depths, thus improving the defects such as local feature omission and overfitting in convolutional neural networks. The simulation experiment was carried out in a balanced public data set and an actual data set. Using the commonly used algorithm SVM as a baseline, the proposed model is compared with ANN, CNN, RF, Bayes and two latest models. Experimentally, the accuracy rate with multiple classifications is 99.5%. The proposed model has the best anomaly detection. And the proposed method outperforms other models in precision, recall, and F1. It is demonstrated that the model is not only efficient in detection, but also robust and robust to different complex environments.