A Robust Outlier Detection Scheme for Collaborative Sensor Networks
L. S. Jayashree, S. S. Arumugam, K.R. Vijayalakshmi · Journal of Digital Information Management · 2007
In-networks, Data Aggregation is usually war- ranted for distributed wireless sensor networks, owing to reliability and energy efficiency reasons. Sensor nodes are usually deployed in unattended and unsafe environ- ments and hence are vulnerable to intentional or unin- tentional damages. Individual nodes are prone to differ- ent type of faults such as hardware faults, crash faults etc and other security vulnerabilities wherein one or more nodes are compromised to produce bogus data so as to confuse the rest of the network in collaborative sensing applications. The availability of constrained resources and the presence of faulty nodes make designing fault toler- ant information aggregation mechanisms in large sensor networks particularly challenging. In our work, we con- sider Byzantine type of faults, which encompasses most of the common sensor node faults (9). Faulty nodes are assumed to send inconsistent and arbitrary values to other nodes during information exchange process. These val- ues are termed as outliers and we use a statistical test called Modified Z-score method to reliably detect and re- move outliers. We show by simulation that the proposed strategy works well for 2 major classes of collaborative sensor network applications viz. (i) Target/ Event detec- tion and (ii) Continuous data gathering. Categories and Subject Descriptors C.2.1(Network Architecture and Design);Wirelss communication: C.4(Performance of Systems);Fault tolerance: E.1(Data Structures); Distributed data structures