Research On DDoS Attack Detection Method Based On Machine Learning In Blockchain Environment

Distributed Denial of Service(DDo S)attack is a kind of network attack,which is often formed by attackers launching a large number of service requests to the target object through various means,which eventually causes the target object to crash,thus triggering further harm.In recent years,emerging Internet technologies such as blockchain,5G network,Internet of Things,and cloud computing have flourished.DDo S attacks in these network environments have the characteristics of low launch cost,difficulty in defense and tracking,and large degree of damage,which are more harmful enough that it caught the attention of industry and academia.Low-performance devices are widely used in each node of the blockchain.Therefore,this paper conducts research on lightweight methods for the difficult problem of DDo S attack detection in the blockchain environment.The main research contents and specific work of this paper are as indicted as below.(1)The principium of DDo S attack and the harm of DDo S attack in these years are analyzed in this paper.The DDo S attack detection methods in recent years are classified and discussed.The development direction of DDo S attack detection in the future blockchain environment is analyzed.At the same time,the principles of gradient boosting machine algorithms including Light Gradient Boosting Machine(Light GBM)and other methods are analyzed,as well as the principles and applications of neural networks and attention mechanisms.(2)A multivariate network flow feature extraction method is proposed to extract the features of network flow,so as to improve the DDo S attack detection performance while ensuring the lightweight input of the detection method.The experimental results show that,compared with the feature groups such as Multi-element Fusion Feature(MEFF)and 6-Feature Group(6-FG),the multivariate network flow feature group proposed in this paper makes the DDo S attack detection results have higher accuracy,lower error rate and missing rate.(3)A lightweight DDo S attack detection method based on Light GBM is proposed for lightweight detection in the blockchain environment.The experimental results show that compared with the detection methods based on Support Vector Machine(SVM),Multilayer Perceptron(MLP)and Deep Neural Networks(DNN),the proposed method higher accuracy,precision and recall,as well as lower error rate and missing rate in the public data set and the blockchain environment simulation experiment.At the same time,it uses a lower cost of computing resources,which can meet the detection requirements of low-performance device nodes in the blockchain.(4)A DDo S attack detection method based on feature-related attention MLP is proposed,which is proposed by combining the feature-related attention mechanism proposed in this paper with MLP.The experimental results on public datasets and the simulation experimental results of blockchain environment show that compared with the methods based on MLP and DNN,the method proposed in this paper has higher accuracy,precision and recall,as well as lower error rate and missing rate in DDo S attack detection.