Research On Anomaly Detection Based On Deep Belief Network

In oder to solve the problem of low detection rate of intrusion detection systems in massive unbalanced data sets,an anomaly detection method based on deep belief network is proposed.Explore the availability and effectiveness of deep belief networks and machine learning algorithms in intrusion detection.Combined with the workflow and related mechanisms of intrusion detection,an anomaly detection framework based on deep belief network is proposed,and other work is carried out based on this framework.Aiming at the serious imbalance of data categories in the existing massive data sets,an anomaly detection method based on synthetic minority samples and deep belief networks is proposed.Firstly,in the data preprocessing module of intrusion detection,the synthesis of minority class technology is used to increase the balance of the data sets,and the improved synthesis of minority class technology increases the minority sample with more minority characteristics.Detection rate for intrusion detection..Secondly,in the data processing module,according to the characteristics of the better nonlinear learning ability of the deep belief network,the three-layer superimposed restricted Boltzmann machine and the single-layer BP neural network are used to reduce the dimensionality of high-dimensional data.Finally,an effective softmax multi-classification method is selected to solve the multi-classification problem of intrusion detection.In the experimental part,the iteration number and activation function of the network structure are discussed,and the random deactivation rate of the node unit is set to enhance the robustness of the network.The validity of the method for synthesizing the minority sample is verified.Experiments on KDD 99 and NSL_KDD datasets demonstrate the effectiveness of the proposed algorithm by comparing it with other algorithms.