Network Intrusion Detection Method Based on FSSAE and GRU

Inhalt:
For the sake of settling this issue that multi classification accuracy of current intrusion determination algorithms for network intrusion is generally not high, in view of the time series characteristics of network intrusion data, a network intrusion determination method of feature selection stacking sparse autoencoder gated recurrent unit (GRU) is suggested. For the sake of settling this issue of wide distribution and strong discreteness of primary attack information, firstly, the data are encoded and normalized, and the XGBoost algorithm and Spearman relevant coefficient are used to choose information characters. Then, the stacking sparse autoencoder is applied to draw features of the information. In the end, GRU completes training of model and realizes the classification. The consequences indicate precision of the suggested approach of the above data sets can reach 99.53%, which is 1.90% higher than that of the same type of SSAE-DNN.