We proposed a new Genetic Programming algorithm termed GPSense for use in wireless sensor networks (WSN). Existing algorithms for pattern recognition and data mining in WSNs work offline, i.e. they query the WSN nodes, bring data to the base-station and the mining is done at the base-station. This increases the latency of decision making, enhances decision communication costs and generally leads to non-local decisions. It is generally believed that this paradigm is more powerefficient since sensor nodes are significantly constrained in terms of computing power (processor speed, low-power batteries, limited-memory). We believe that a distributed data mining approach can be evolved where small-footprint mining algorithms can be developed to work on the sensor-nodes thereby improving the current state-of-the-art. GPSense is the first of such in-network data mining frameworks and has the following desirable characteristics:
It is designed to work as a distributed algorithm that co-ordinates and exchanges genetic material with collaborating nodes.
It is aware of the resource constraints at node level with its footprint being consistently smaller than that of the sensor node's processing capabilities.
Its localized nature - enables the entire WSN to make decisions at the nodelevel instead of aggregating results from long-running continuous queries to the root node for eventual filtering.
In this thesis we describe the GPSense framework and demonstrate its utility based on the results we obtained on a test-bed WSN.
Library of Congress Subject Headings
Wireless sensor networks--Data processing; Wireless LANs--Data processing; Genetic algorithms
Computer Science (MS)
Department, Program, or Center
Computer Science (GCCIS)
Rajendra K. Raj
Soni, Soujanya, "GPSense: an algorithmic framework for intelligent sensing at node level in WSN" (2007). Thesis. Rochester Institute of Technology. Accessed from
RIT – Main Campus