Research On Network Intrusion Detection Technology Based On Deep Learnin

Network-Based Intrusion Detection System(NIDS)is an important shield to protect network security after firewall technology,and has a great research value and significance in high-speed network development nowadays.However,the accuracy and speed of network intrusion detection in current algorithms have not been compatible well,which greatly limits the application and development of network intrusion detection.Moreover,the number of different kinds of data is very unequal in the public dataset,which easily leads to poor detection accuracy for classes with low sample counts.Therefore,detailed research is conducted in this paper in terms of both improving the efficiency of network intrusion detection and ameliorating the imbalance of the dataset.The main research contents and results are as follows:(1)To address the problem that existing machine learning algorithms are fast but low in accurate,while deep learning algorithms have long training time and do not balance time and accuracy well.In this paper,we propose an efficient network intrusion detection model based on the Light GBM classifier.In order to reduce the dimensionality of the preprocessed data and improve the classification efficiency of the classifier,a feature selection algorithm is designed on the basis of principal component analysis and Information Gain method.The feature selection algorithm combined with the classifier can make the model achieve better classification results in a short time.The experiments are conducted with the UNSW-NB15 dataset,and the results of binary classification and multi-classification show that the method in this paper can effectively reduce the classification time and improve the detection accuracy and recall rate of different types of data in the dataset.(2)To address the problem of low classification accuracy of few class samples caused by unbalanced datasets,an intrusion detection algorithm DBEA that combines integrated classification with changing the number of samples is proposed to solve the classification problem caused by unbalanced datasets.The DBEA method builds a deep autoencoder network to reduce the dimensionality of the encoded data.And the AGM method is designed based on ADASYN and GMM algorithm,which can modify the number of partial classes in the dataset and improve the dataset imbalance.Finally,an ensemble classifier is designed based on Gaussian Na?ve Bayes,C4.5 and Random Forest algorithm,and the classification effects of multiple classifiers are combined by improved voting method.The experiments are conducted using NSL-KDD and CICIDS2017 dataset.The results show that the accuracy of the algorithm in this paper is improved in detecting different attack samples in the dataset by comparing multiple algorithms to deal with the data imbalance problem.