Network Intrusion Detection Based On Integrated Algorithms And DBN Network

Nowadays,our life is closely related to the Internet,the internet brings convenience and also has hidden dangers.In order to eliminate these hidden dangers,people put forward network intrusion detection technology.Especially in recent decades,the unprecedented development of machine learning and neural network has also promoted the vigorous development of network intrusion detection technology.Network intrusion detection refers to the technology of real-time detection of network state in order to prevent network intrusion.It is an active defense technology,its existence will not affect the Normal operation of the network.The importance of network intrusion detection technology is increasing,If firewall is a barrier to prevent intrusion,then network intrusion detection is the second barrier to prevent network intrusion.In this paper,in order to study the problem of network intrusion detection,a network intrusion detection model is established by using a better integrated algorithm in machine learning algorithm.Specific work and main research contents are as follows:1.Preprocessing of network intrusion detection data.Aiming at the problem of network intrusion detection,this paper analyses and compares two data sets Kddcup99 and NSL_KDD commonly used in the world.Then the NSL_KDD data set is preprocessed in three stages:the stage of turn labels into numerical phases,the stage of converting character formal attributes to numeric formal attributes and the stage of normalization.2.Research on Network Intrusion Detection Based on integrated algorithm.Firstly,this paper uses the integrated algorithm random forest to model and process the network intrusion detection problem.Secondly,the integrated LightGBM algorithm is used to model the network intrusion detection,and Bayesian optimization is used to optimize the parameters.Then two different models are built according to the nature of CatBoost algorithm when using integrated algorithm CatBoost to model network intrusion detection problems: the network Intrusion Detection Model Based on CatBoost Self-Processing Category Characteristics and the network Intrusion Detection Model Based on Integrated Algorithms CatBoost after Converting to Numeric Data.Then we compare the detection results of the three integrated algorithms and compare them with those of other machine learning algorithms.And SMOTE algorithm is used to process a fewdata sets.Finally,the Stacking model fusion technology is used to fuse several integration algorithms with a better two-tier Stacking structure to achieve better detection results.3.Research on Dimension Reduction Based on DBN Network Features.When using NSL_KDD data set for network intrusion detection research,the data set will be processed first,and then processed into data sets with 121 features.The feature may be redundant,so in order to further improve the detection effect,we focus on feature dimensionality reduction.DBN network is used to extract the feature nonlinearity,and then the data set after the new feature is input to the integration LightGBM algorithm.In order to verify the effectiveness of the proposed dimensionality reduction method using DBN network first,It is compared with the traditional feature extraction method PCA technology and only use LightGBM algorithm at the same time.The experimental results show that the detection effect of DBN-LightGBM is the best.So it is also verified that the detection technology of DBN-LightGBM proposed in this paper comprehensively utilizes the advantages of DBN's non-linear feature extraction and the excellent classification and prediction performance of the integrated LightGBM algorithm.