Research On Network Intrusion Detection Based On Improved Auto-Encoder

Efficient intrusion detection algorithm is the core technology of intrusion detection system.The traditional network intrusion detection algorithm has low detection accuracy and high false positive rate,so it can not effectively detect the increasingly complex network intrusion behavior.Therefore,it is of great significance to improve the performance of intrusion detection algorithm.As a deep learning algorithm,the Auto-encoder can effectively reduce the feature dimension of the dataset and guarantee the integrity of the original dataset.The application of Auto-encoder in network intrusion detection system can extract effective information from high-dimensional data,so as to improve the performance of intrusion detection system.Based on the above principles,this paper proposes an intrusion detection model based on improved Auto-encoder.Based on the study of the Auto-encoder algorithm,this paper proposes an intrusion detection model based on stack sparse Auto-encoder combined with probabilistic neural network(SSAE-PNN),aiming at the disadvantages of slow convergence speed and high false positives.This model can extract the complex relationship between features,which can not only reduce the feature dimension,but also learn the hidden and deep features.The classifier of this model has fast convergence speed and high classification accuracy,which reduces the training time and saves the time cost.The experiment shows that compared with other detection methods,the intrusion detection model based on SSAE-PNN has higher detection accuracy and lower false alarm rate,which is effective for detecting network abnormal attacks.Aiming at the shortcoming of weak anti-interference ability of SSAE-PNN model,this paper proposes a stack-contraction Auto-encoders combined with probabilistic neural network(SCAE-PNN)intrusion detection model.This model has both strong feature extraction ability and anti-jamming ability,which overcomes the weakness of SSAE-PNN based intrusion detection model.In addition,the UNSW-NB15 dataset was used to train and test the SCAE-PNN model.The UNSW-NB15 dataset contains newer and more comprehensive attack types than the NSL-KDD dataset.The experimental results show that: compared with other Auto-encoder models,the accuracy of this method is improved by about 3%,and thefalse alarm rate is reduced by about 2%.