Research On IoT Anomaly Detection Model Based On Stacked Autoencoder And FSVM

The rapid development and wide application of computer technology and Internet technology have promoted the development and popularity of Io T technology.And with the continuous development and application of new biological networking technology,the emergence of Io T intrusion phenomenon has become more and more frequent.Intrusion detection,as a means of network information defense,has always been the focus of the information security field,and it has always played an important role in the protection of information security.However,in the environment of Io T,there are large-scale and more complex characteristics of data in the network environment of Io T,which brings new challenges for the rule learning and data detection of intrusion detection model.The paper combines deep learning technology and classification technology,and introduces it into intrusion detection,and proposes a new intrusion detection algorithm based on stack autoencoder and fast support vector machine(SAE-FSVM)to detect abnormal network data.Support vector machine(SVM)is a supervised learning model with low risk and high generalization ability.With limited sample data,it can obtain a structure with a small error through training,so it is widely used in the field of intrusion detection.However,as a small-sample learning model,its learning ability for large-scale high-dimensional samples is weak,and the learning effect is poor.Large-scale,high-dimensional network sample data often affect the learning speed and learning rate of the SVM model.Therefore,the main content of this paper is as follows:(1)For high-dimensional network data,the algorithm uses stacked autoencoder(SAE)for feature extraction,and abstractly maps high-dimensional network data layers to low-dimensional space through deep structure,so as to obtain deeper and more Abstract,more advanced feature representation.Effectively reduce the impact of irrelevant feature information and repeated feature information on SVM training,and improve the learning speed of SVM.(2)In view of the impact of large-scale sample data on SVM,There are two sample size reduction algorithms to be proposed for reducing the input samples of SVM in this article.According to the training characteristics of SVM,most of the non-support vectors are calculated and filtered,which reduces the training burden of SVM and improves the speed of SVM training.This paper evaluates the abnormal data detection experiment based on the KDD CUP99 data set,and verifies that the SAE-FSVM model can not only effectively improve the accuracy of the SVM classifier,but also greatly reduce the SVM performance under large-scale and high-dimensional sample data.Training time.Experiments show that the algorithm in this paper can effectively reduce the impact of large-scale,high-dimensional network data on SVM classification.