Network Intrusion Detection Model Based on CNN and GRU

A network intrusion detection model that fuses a convolutional neural and gated recurrent unit is proposed to address the problems associated with low accuracy of existing models for multiple classification intrusions class imbalance data detection. In this model, hybrid sampling algorithm combining Adaptive Synthetic Sampling (ADASYN) Repeated Edited nearest neighbors (RENN) used sample processing solve problem positive negative in original dataset. The feature selection carried out by Random Forest Pearson correlation analysis redundancy. Then, spatial features are extracted using network, further fusing Averagepooling Maxpooling, attention mechanism assign different weights features, thus reducing overhead improving performance. At same time, Gated Recurrent Unit (GRU) extract long-distance dependent information achieve comprehensive effective learning. Finally, softmax function classification. evaluated based on UNSW_NB15, NSL-KDD, CIC-IDS2017 datasets, experimental results show reaches 86.25%, 99.69%, 99.65%, which 1.95%, 0.47% 0.12% higher than type CNN-GRU, can well.