Study On Algorithms For Outlier Detection In Wireless Sensor Networks

Wireless sensor networks(WSN). integrating of wireless communications technology, embedded computing technology, distributed information processing technology and sensor technology, can cooperate and run real time detection, sensing and collecting information of the phenomenon, and then process, send it to the end user. It has broad application prospects in military monitoring, environmental monitoring, medical monitoring, building monitoring and space probing etc.Recently, outlier detection in wireless sensor networks is becoming one of the hottest research areas. However the existing outlier algorithms in WSN are of some disadvantages such as lower detection precision, higher communication complexity and computational complexity due to not enough consideration of the spatial-temporal correlation of data and the characteristic of distribution networks. Researching the outlier detection algorithm with high efficiency become the focus of this article. The contribution of the paper is as follows:(1)Analyzing the existing outlier detection algorithm, and specify its advantages and disadvantages, then proposed the idea that making full use of the spatial-temporal correlation to outlier detection in order to improve the detection accuracy,reduce communication.(2) For low-dimensional data, proposed a novel distributed on-line outlier detection algorithm based on spatio-temporal correlation OTOD(Online Three-phase Outlier Detection algorithm). In each sensor node, using sliding window technique generates a set of candidate outliers based time-correlated sensor readings, and using filtering technology generates a set of local outliers based spatial neighborhood. Ultimately, in sink sensor node, collecting whole local outliers in all nodes obtains the set of global outliers according to the outlying degree. Using spatial and temporal correlation improves the detection accuracy, and using distributed computing reduces the amount of communication and computation. Theoretical analysis and experimental results show that OTOD algorithm outperforms the other algorithms in user-dependency, detection accuracy and efficiency.(3) For high-dimensional data, proposed the adaptive hyperellipsoidal SVM-based outlier detection algorithm. Hyperellipsoidal SVM over the drawback that only for spherical distribution of hypersphical SVM.To meet the concept drift of data stream.Introduce a new model update mechanism, by judging the new data and the earliest entry data deviate from the classification model to determine the new data whether impacts the classification model, so to determine whether update the model. And in order to improve the precision, quantify the spatial characteristics of data. Theory analysis and experimental results show, compared with the other exsiting algorithms, this algorithm improve the detection accuracy.