Abstract: A WSN (Wireless Sensor Network) is an ad hoc network of sensor nodes to monitor natural or man-made conditions and to cooperatively pass their data through the network to a main location. The thorough debugging of WSN applications before actual deployment can save a lot of time and manual effort. The existing WSN applications look short and simple, but they frequently encounter failures due to various software bugs. The main reason is that they are executed with a complicated event-driven concurrency model. WSN debuggers based on symptom mining are one of the efficient methods for identifying potential transient bugs in WSN applications. The design of such debuggers is based on a key observation that outliers in the normal behavioral patterns of a WSN system indicate transient bugs. The symptom mining debuggers applies a customized outlier detection algorithm to quickly identify and rank abnormal intervals. The scope of this paper is to implement the design of a symptom mining debugger and to conduct a comparative study of the various outlier detection algorithms in finding the potential transient bugs of a few WSN applications. The paper has identified the optimum features of an outlier detection algorithm to be used in a symptom mining debugger. OCSVM and Orca were selected because of their unsupervised nature and ability to deal with homogeneous multivariate continuous attributes. The merits and demerits of the selected outlier detection algorithms were also identified after performing an execution time and memory overhead analysis.
Keywords: WSN, OCSVM, Sentomist