Outlier Detection Technique On Uncertain Sensing Data

As the development of techniques on micro electronic components and wireless communications, Wireless sensor networks are applied in various fields such as environment monitoring, traffic monitoring and forest fire defending. As one of the most important applications, outlier detection becomes more and more important. However, because of the traits such as low accuracy, limited hardware resources, fragile anti-disturbance, data is always collected with uncertainty, meanwhile, sensor nodes would produce errors as a result of environment noise, hardware disturbance, environment temperature, lost of energy and hardware failure. Data with errors will have a disgusting influence on accuracy of the query results, what is more, it interfere the normal work of WSN so that waste lot of unnecessary loss. So an outlier detection technique with high accuracy and efficiency is very desirable in practical applications.To solve the problems mentioned above, let sensing data in wireless sensor networks be the research center, we thoroughly analysis the characteristic of sensor data with uncertainty, present the uncertain data model in WSN, base on this, we propose the definition of distance-based occasion outlier and the conception of time-based continuous outlier in wireless sensor networks. According to the definition of occasion outlier, we propose a pruning algorithm named GPA, which can significantly improve the efficiency for detecting occasion outlier. By using the structure and some special properties of Grid, three underlying pruning strategies are found for efficiently pruning uncertain tuples on data snapshot. In accordance with the continuous outlier, we convert the states of tuples in different snapshots for the same node to 0-1 sequence, and use a sub-sequence matching method for finding the continuous outlier in WSN.In the experiment, simulate uncertain dataset is used for evaluating the performance of GPA and sub-sequence matching method. For GPA, we firstly evaluate the running time and pruning ratio by varying the parameters of occasion outlier; secondly, we evaluate the scalability of GPA by varying the distribution of sensor data and the scale of WSN. The experimental results show the powerful pruning ability and good scalability of GPA, which can response occasion outlier fast in large scale of WSN. For sub-sequence matching method, the feasibility is demonstrated by testing the recall and precision of continuous outlier, under the condition that parameters are well set.