Efficient Density Based Techniques for Anomalous Data Detection in Wireless Sensor Networks
N. Chitradevi, Vadivel Palanisamy, K. Baskaran, K. Swathithya · 2013
Recently sensor networks have found many applications which address data reliability as an important issue and demand accurate prediction of phenomena. In wireless sensor networks data integrity is affected by the harsh environmental conditions, interferences in wireless medium, low quality sensors and remote unattended operation nature. Due to these conditions, the data sensed by the sensors get corrupted resulting in outliers causing the observed phenomena to deviate from actual phenomena. Hence outlier detection forms an important research topic. To address the problem of unsupervised outlier detection in wireless sensor networks, we develop two density-based outlier detection techniques to discover local outliers using k-distance neighborhood based local outlier factor (LOF) formulation. Our proposed mechanisms are designed in two aspects: minimizing the computational time for LOF calculation to cope with stringent resource constrains of sensor nodes and enabling the detection of outliers lying in dense and sparse clusters. Using real and simulation dataset we demonstrate that our first approach DBOD_MSS needs less computational time for LOF determination but still provide equivalent detection accuracy as of existing approach and the second approach DBOD_PMSS successfully detects outlier appearing as small groups which are not identified with existing approach.