Research On Intrusion Detection Based On Deep Learning And Twin Support Vector Machine

In today's network environment,a large amount of data with high complexity is produced every day.In order to solve the problems of improper extraction of data features,low accuracy and slow speed caused by these high-dimensional data for intrusion detection,a DBN-TSVM-5 intrusion detection model based on deep belief network and twin support vector machine was proposed.Firstly,the character mapping and data normalization of intrusion detection data set KDDCUP99 were carried out to obtain the standard data set that can be used in intrusion experiments.Then the deep belief network model made by the five-layer restricted boltzmann unit was used to extract the features of the standard data and obtain the intrusion data with low dimension and the characteristics of the original samples.A multiclassification TSVM-5 classifier was proposed to detect and identify five kinds of intrusion data.Finally,a simulation experiment was established to verify that the dbn-tsvm-5 model is an effective method.Feature data is directly related to the results of intrusion detection experiments.Therefore,selecting effective feature learning methods is an important step that cannot be ignored in intrusion detection.Through comparison of feature dimension reduction experiments,the results show that the detection accuracy of deep belief networks is 25.85%,25.77%,and 1.19% higher than that of principal component analysis,linear discriminant analysis and T-distributed stochastic neighbor embedding.The performance of support vector machines and paired twin support vector machines in intrusion detection was analyzed.Experiments show that twin support vector machines have advantages over support vector machines when processing large-scale data sets,and the detection accuracy rate has increased by an average of 1.38%.A TSVM-5 algorithm was proposed on the basis of twin support vector machine.The experiment shows that the algorithm to identify the five types of data in KDDCUP99,which has an average accuracy improvement of 5.37% over the traditional twin support vector machine detection.The proposed DBN-TSVM-5 model has an average detection accuracy improvement of 2.52% over TSVM-5.Comprehensive experimental results show that the DBN-TSVM-5 model improves the accuracy of intrusion detection and reduces the false positive rate.Figure 31;Table 17;Reference 61...