Research On Intrusion Detection Methods For The Internet Of Things Based On Machine Learning

The Internet of Things(Io T)is facing serious risks in information security because of its wide range of applications and massive unattended nodes.The nodes of the sensing layer of Io T have the characteristics of low computing power,low energy consumption and insufficient storage space.The existing intrusion detection methods are difficult to adapt to limited resources and constantly changing environment.In view of the characteristics and security requirements of the sensing layer of Io T,this thesis specially conducts an in-depth and systematic study on the Io T intrusion detection method which based on machine learning,and discusses the lightweight and intelligent Io T intrusion detection method.The main contributions include:Firstly,The Least Squares Support Vector Machine(LSSVM)model has fast training speed and high accuracy but lacks sparsity.a scheme which intrusion detection classifier of Io T based on sparse LSSVM is established to improve classification efficiency and accuracy.At the same time,reduce the occupancy of computing resource.Secondly,in views of the problem that the initial data set has a large number of samples and it is difficult to conduct model training in a resource-constrained environment,a data sparse method is proposed to reduce the sparse support vector before model training.Among them,for the clustering speed is slow and it is easy to fall into the local optimal solution of the K-means clustering algorithm,The improved simulated annealing algorithm is used to optimize the initial cluster center point to accelerate the clustering.The influence of noise on the classification effect is proposed.The pauta criterion is used in the clustering cluster to judge the noise point and denoise.The Euclidean distance sample selection method is introduced to improve the efficiency.The nearest and farthest of the heterogeneous samples are selected quickly and effectively by the cluster center point.The simulation test results show that the sparse rate of the improved data sparse method is 50.66%,and the model accuracy rate is 97.5%.Thirdly,in view of the problem that the storage space of the Io T node is limited and the storage space occupied by the support vector library is large after the data is sparse,a kernel matrix sparse method is proposed.The method creatively combines the filtering feature selection and the pruning method,selects the column in the kernel matrix that plays an important role in classification,preserves the corresponding support vector of the column,and transfers the information of the non-support vector to ensure the model is prepared.Ability.The simulation test results show that the support vector is reduced from 2467 to 147 by kernel matrix sparse,the sparse rate is 94.04%,and the classification accuracy is 96.75%.In short,based on the machine learning method,the least squares support vector machine classifier realized by data sparse and kernel matrix sparse can effectively improve the detection accuracy while ensuring the lightweight and intelligent Io T intrusion detection method.