Sensor Network & Weather Data Stream Mining
Hakilo Ahmed Sabit, Adnan Al‐Anbuky · 2011
Interest in environmental studies and awareness reflects sustainable growth. This in effect results in growing volume and speed of environmental data acquired from field sensors and systems. Real world environmental spatial data have the form of being vast, highdimensional, raw, sparse, and lends itself to being dynamically streamed. Data with such characteristics calls for handling approaches and methodologies different from traditional ones in support for valuable environmental information. This paper proposes an efficient machine learning algorithm with less than 10% error bounds to extract bushfire hazard information from high arrival rate weather data streams using the resource-constrained wireless sensor network (WSN). The aim is to provide high spatial and temporal resolution bushfire hazard report. This could be achieved by employing dense multi-point and low cost weather sensors. The bushfire hazard prediction model of the Canadian fire weather index (FWI) system is implemented via in-network distributed computing. Simulation results indicate the potential of this approach to efficiently handle weather data stream mining with significant improvement in both temporal and spatial resolution.