Research On Intrusion Detection Based On Machine Learning

Albeit having provided great convenience in the lives and works of the people recently,the rapid development of the computer internet industry has been riddled with several network security related threats,various types of vulnerabilities,viruses,and malicious attacks leading to heavy losses to the internet assets.As the trend of network threats marches ahead,the defense technologies and means of information security have been constantly iterating and updating.Although the existing tools have certain protection effects against the network threats,they are unable to handle the increasingly serious network threats,while the intrusion detection technology has been widely used because of active identification and defense in the face of massive network intrusion.Through a plethora of existing research results related to intrusion detection,it was found that the essence of the applicable intrusion detection technology was to classify the network attacks,nevertheless,most of the existing research had ignored the problem of a wide variety and range of attacks along with the uneven distribution of the number of attacks on the network.The rapid developments in the field of machine learning have created new opportunities for intrusion detection techniques.In this paper,the machine learning algorithms have been applied on intrusion detection to investigate the effectiveness and feasibility of the machine learning related data processing methods and classification algorithms.Hence,this paper addresses the problem of uneven distribution of the different types and numbers of attacks on the network by carrying out the following works.(1)An intrusion detection method has been proposed that integrates the Deep Belief Network(DBN)and the Least-Squares Support-Vector Machines(LSSVM).The DBN could effectively implement the dimensionality reduction when dealing with large quantum of high-dimensional data,which makes the model more computationally efficient and better.The LSSVM in integrated learning has been found to have good classification training ability and could be integrated with other models to achieve good desired results,nevertheless,dealing with high-dimensional feature data would be highly challenging.Hence,an intrusion detection fusion scheme that mainly implements the data feature downscaling and improves the classification efficiency has been proposed.After verification,the DBN-LSSVMbased intrusion detection method has proved to be an effective detection method with good performance and more stability.(2)The traditional DBN tends to lose a part of the original information in the process of extracting high-dimensional features of the data,resulting in the reduction in the correlation between the original features of the data or the loss of certain features that were beneficial for subsequent classification.To address this problem,an improved DBN(IDBN)model has been proposed by merging the original input data into the layer-by-layer training process several times to optimize the extraction of the original features and the feature correlation.Whereas,by applying it to intrusion detection,it has been demonstrated that the model could effectively improve the effects of the model feature extraction.(3)Addressing the problem of uneven distribution of the various types of attacks in the network attack traffic and the poor performance of existing intrusion detection models in detecting rare attack types,this paper proposes an integration of the Generative Adversarial Networks(GAN),Improved Deep Belief Network,and the e Xtreme Gradient Boosting(XGBoost)algorithm for effective and efficient intrusion detection.The original intrusion detection dataset has been expanded with the rare attack type samples using GAN and the expanded dataset has been downscaled using IDBN to maximize the retention of the original features,followed by the classification of the intrusion detection dataset using the parameter-tuned XGBoost model.Experiments have illustrated that the GAN-IDBN-XGBoost-based intrusion detection method performed well in the binary classification experiments and significantly improved the detection and accuracy rate in the detection of a few specific classes of attacks.