Research On Multi-Class Intrusion Detection Based On Deep Learning

The rapid development of the Internet has pushed human society into a highly information-based network society,the Internet has penetrated into all aspects of people's lives.While the Internet brings convenience to people's lives,it also faces many security risks.Traditional intrusion protection methods can no longer be well adapted to the current high-speed and large-scale network environment.Some researchers have applied artificial intelligence technology to the field of intrusion detection and have achieved good results.At present,intrusion detection technology is facing many challenges.Improving the performance and efficiency of intrusion detection technology is very important to the security of individuals,companies,and the country.This thesis first studies the common detection algorithms that apply deep learning technology in the field of intrusion detection.Aiming at the problem of poor detection performance caused by the single current model,this thesis proposes an anomaly detection method based on the CNN?Bi LSTM network.The data is preprocessed from captured network data.To reduce data redundancy,this thesis uses feature selection algorithm to select important attributes.We use processed data as input to subsequent modules,and then connect CNN and Bi LSTM modules to extract the spatio-temporal features of the data.In order to express useful input information,the self-attention mechanism is used,which assigns different weights to the fused features.Gated Recurrent Unit is used to train the model.Finally the softmax function is used for classification.Secondly,in view of the data imbalance problem that exists widely in various fields,this thesis proposes a data imbalance intrusion detection method based on mixed sampling.The model first reduces the problem of data imbalance from the feature level and the data preprocessing level.The feature level mainly reduces data redundancy through feature selection,and the preprocessing level uses hybrid sampling technology to reduce data imbalance.Aiming at the decision boundary samples and safety samples in minority samples are over-sampled based on the threshold algorithm.For the majority samples,K-means is used to cluster them,and random under-sampling is performed proportionally according to the clustering results.For the ordinary classes that do not need sampling are not processed.Then we build multi-class balanced data set,the sample obtained by mixed sampling is used as the input of each heterogeneous classifier.Then multiple heterogeneous classifiers are constructed.Finally,combined with each training sub-classifier,the voting mechanism is adopted to determine the result of the decision to check the classification effect.Finally,the method proposed in this thesis is evaluated and analyzed on the UNSW?NB15 data set.The experimental results show that the proposed method is superior to the existing detection system in terms of comprehensive performance,and improves the detection efficiency,correct rate and accuracy rate.In terms of unbalanced data,the hybrid sampling method proposed in this thesis can effectively improve the classification quality,which illustrates the effectiveness of the sampling method.