Outlier Detection in Wireless Sensor Networks Basedon OPTICS Method for Events and Errors Identification (original) (raw)

Wireless Sensor Network is composed of small, low cost, low energy, and multifunctional sensors. In addition, this network could have scalability, topology, synchronization, radio-coverage, safety and security constraints. Therefore, our challenge is to classify data into normal and abnormal measurements using outlier detection methods. This paper explore the density-based method Ordering Points to Identify the Clustering Structure. Proposed detector applies an auto-configuration of parameters without previous known environmental conditions. It also extracts hierarchical clusters that serve in a postprocessing treatment for classification of data into errors and events. Performance is examined within a real and synthetic databases from Intel Berkeley Research lab. Results demonstrate that our proposed process analyzes data of this network with an average equal to 81% of outlier detection rate, 74% of precision rate and only 2% of false alarms rate that it is very low compared to other methods.