Distributed Anomaly Detection Algorithms

Anomaly detection has attracted much attention in recent years since it plays a crucial role in many domains.Anomaly detection is a technology that finds the data,which does not conform to a well-defined normal data,through data analysis.In practice,data used for anomaly detection can be distributively collected via sensor networks.Besides,as the data usually arrives at the nodes sequentially,online detection method that can process streaming data is preferred.In this paper,we focus on distributed online anomaly detection algorithms.For the study of two main kinds of learning problems in the distributed environment,namely,the unsupervised learning and the su-pervised learning,we propose two distributed online algorithms for these learning problems based on the PCA and OCSVM,respectively.Firstly,for unsupervised learning problems,we study the application of the PCA for anomaly detection and propose a distributed PCA objective function.In order to achieve the online learning,we modify the objective function through an appropriate approximation.Then,for supervised learning problems,we study the mechanism of OCSVM,and propose a distributed cost function.To obtain a decentralized implementation without trans-mitting the original data,we replace the kernel function by a random approximate function and minimize the cost function by stochastic gradient descent(SGD),Besides,to find an appropriate approximate dimension,we propose a sparse distributed online one-class support vector machine algor:ithm.Some experiments are performed to show the effectiveness of the proposed algorithms.Experimental results on both synthetic and real datasets reveal that both of the proposed algorithms achieve low misdetection rates and high true positive rates.Compared with other state-of-the-art anomaly detection methods,the proposed distributed algorithms have good anomaly detection per-formance.