Abstract
In order to ensure network security, a multiple classification detection model based on DeepFM framework is constructed to improve the accuracy of network traffic anomaly detection. Firstly, the FM module was adopted to make the model self-build cross features, and the A-ResNext model was introduced into the DNN module to perform group convolution operation on features. Then, it is integrated with self-attention mechanism to reduce redundant features. Finally, the network classifier is used to realize the multiple classification detection of traffic anomalies. Experimental results showed that the accuracy and F1 value of the proposed multiple classification model in malicious traffic detection task are 82.18% and 78.95%, respectively, which are higher than those of the traditional classification detection model. It can be seen that the proposed model is feasible and effective to accurately detect the abnormal state of network traffic.