Based ontwice decision for deep learning intrusion detection model

Aiming at the intrusion detection method of deep neural network, there are two problems of data imbalance and feature redundancy in the training process, resulting in low detection rate and high false positive rate. Based on Twice Decision Deep Learning model(TDDL) is proposed: The model is a combination of Deep Stack Autoencoder(DSAE) and neural network, including two-stage feature learning, in which the first stage uses DSAE to compress features and add probability value features that distinguish abnormal data, and the second stage uses neural networks(NN) receives the characteristics of the first stage and trains, thereby reducing the feature redundancy and balance bias on normal data to improve the detection effect. The experimental results of KDDCUP99 dataset show that the model can effectively improve the effect of deep neural network on feature detection of intrusion detection data, so that it has higher accuracy and lower false positive rate.