Research On Privacy-preserving Anomaly Detection For Blockchain

With the rapid development of blockchain technology in digital currency,the economic value of blockchain system is rising.Huge economic interests induce criminals to attack blockchain system by various means of attack.Attacks such as theft,ransomware,fraud,etc.emerge in endlessly in digital currency,which makes the security situation of blockchain system more severe.The security of blockchain system has become a key problem that restricts its development,and various security threats and potential attacks it faces have been widely concerned by academic circles.The ultimate goal of attacks on blockchain system is to steal users’ digital currency in order to gain huge economic benefits,and abnormal transactions in blockchain often occur in the process of attacks.If we can study the abnormal transaction detection,we can effectively discover the attack behavior of criminals,give early warning of possible attacks or illegal transactions,and effectively guarantee the security of blockchain system,which has important theoretical value and practical significance.In this paper,the abnormal transaction detection of blockchain is studied by machine learning method,and the privacy protection for sensitive transaction abnormal detection is further studied,and the following results are obtained:1.A blockchain abnormal transaction detection scheme based on supervised machine learning is proposed for bitcoin theft.At present,the research on abnormal transaction detection of blockchain is divided into general detection research and special detection research,but the special detection for bitcoin theft with great destructive power is still lacking.In view of this,we propose an abnormal transaction detection scheme for bitcoin theft based on supervised machine learning algorithm.Firstly,the features of bitcoin public transaction data set are extracted according to the characteristics of theft transaction,and then five supervised methods KNN,SVM,RF,Adaboost,MLP,three unsupervised methods LOF,OCSVM and Mahalanobis distance-based method are used to detect the extracted feature data.On this basis,the training data are equalized to improve the detection effect of supervised methods.Experiments show that the supervised method before and after equalization has achieved good detection effect on the features we extracted,and the detection effect has been further improved after equalization.Before and after equalization,KNN,RF and Adaboost have good detection effects,with F1 values above 80%,among which RF performs best,and the effect is improved after equalization.The recall rate,precision and F1 value of RF before equalization are 92.4%,98.1% and 95.2% respectively,and after equalization,the recall rate,precision and F1 value are 95.9%,95.9% and 95.9% respectively.2.A block chain abnormal transaction detection scheme with privacy protection function is proposed.At present,the regulatory demand for consortium blockchain is becoming more and more prominent,but the current research on abnormal transaction detection in consortium blockchain is relatively rare,and few related researches mainly focus on setting up abnormal detection mechanism within consortium blockchain,which does not meet the requirements of external supervision.Because of the privacy of consortium blockchain data,this paper proposes a scheme of detecting abnormal transactions in blockchain under the condition of privacy protection to realize external supervision of consortium blockchain.In this scheme,the consortium blockchain accounting node randomizes the transaction data by matrix multiplication and sends it to the cloud server.The cloud server performs anomaly detection based on KNN and feeds back the detection results to the consortium blockchain accounting node for verification.We have carried out theoretical analysis and experimental simulation on the correctness,security and performance of the detection scheme.Experimental results show that the scheme has little influence on the efficiency of consortium blockchain and has good detection effect.The best recall rate,precision and F1 value of KNN can reach 85.3%,87.7% and 86.5%.3.A strong privacy protection scheme for detecting abnormal transactions in blockchain is proposed.Using matrix randomization to protect the privacy of transaction data has high efficiency,but its security depends on matrix inversion.After collecting enough matrix multiplication randomization transaction features,attackers may destroy the privacy of transaction features by solving linear equations,which is not suitable for application scenarios with strong demand for privacy protection.In view of the above problems,a blockchain abnormal transaction detection scheme with strong privacy protection is proposed.In this scheme,firstly,the extracted transaction features are discretized into positive integers,then the transaction features are encrypted by Paillier encryption algorithm with additive homomorphism,and then the encrypted transaction features are randomized by matrix multiplication to achieve stronger privacy protection.We have carried out theoretical analysis and experimental simulation on the correctness,security and performance of the detection scheme.The results show that the scheme not only has stronger privacy protection characteristics,but also has higher execution efficiency and detection effect.In this paper,the abnormal transaction detection of blockchain is studied in three parts.First of all,this paper improves the characteristics of the public chain digital currency Bitcoin abnormal events and uses supervised methods to detect Bitcoin theft events.Secondly,according to the privacy protection requirements of abnormal transaction detection in consortium blockchain,the transaction characteristics are randomized by matrix multiplication and transmitted to the cloud server for detection,so as to realize abnormal detection under privacy protection.Finally,for the application scenarios with high transaction sensitivity,this paper uses homomorphic encryption technology to enhance the privacy protection effect based on the above work.To sum up,the research results of this paper provide theoretical basis and important reference for the research of abnormal transaction detection in blockchain under privacy protection.