1D CNN for Feature Reconstruction on Network Threat Detection
Yanping Xu, Xia Zhang, Tingcong Ye, Zhenliang Qiu, Lingjun Zhang, Hua Zhang, Yifan Wu · 2021
Machine learning algorithms for building network threat detection model are regarded as effective methods. Some big network security data has 1D characteristics. And 1D CNN can deal with 1D signal data well. Therefore, in this paper, we propose a network threat detection model based on 1D CNN and shallow machine learning algorithms. Firstly, 1D CNN deep model is constructed, including input layer, convolution layer, pooling layer, full connection layer, softmax layer and output layer. Combining the layers nonlinear learning, reconstructed feature data is generated, which reduces the dimension of the features comparing with that of original data. Secondly, the reconstructed data is input into the shallow machine learning algorithms to classify the network security data and detect the abnormal behavior in the security dataset. The experimental results show that the classification performance of 1D CNN is outstanding comparing with other dimension reduction algorithms.