Research on Malicious Traffic Detection Methods Based on Deep Learning

Abstract

The number of malicious traffic in the real network is growing rapidly, such as common DoS attacks, web attacks, DDoS attacks, and so on. These attacks may cause huge economic losses to enterprises, countries, and individuals. Traditional machine learning methods are hardly meeting the accuracy and real-time requirements of malicious traffic identification technology in the face of huge and multi-dimensional online data. Large-scale data can provides training data for deep learning, and deep learning can directly obtain advanced features from the data. Therefore, the malicious traffic identification technology based on a convolutional neural network (CNN) is adopted. The outliers and useless features in the data set are removed by preprocessing, and then the model is obtained by multiple training and optimization of the convolutional neural network. The experimental results show that CNN performs well in the malicious traffic detection tasks, and the metric F1 values reach 95%, which is better than the deep learning models such as RNN, and CRNN.