On-board Mining of Data Streams in Sensor Networks

Abstract

Data streams are generated in large quantities and at rapid rates from sensor networks that typically monitor environmental conditions, traffic conditions and weather conditions among others. A significant challenge in sensor networks is the analysis of the vast amounts of data that are rapidly generated and transmitted through sensing. Given that wired communication is infeasible in the environmental situations outlined earlier, the current method for communicating this data for analysis is through satellite channels. Satellite communication is exorbitantly expensive. In order to address this issue, we propose a strategy for on-board mining of data streams in a resource-constrained environment. We have developed a novel approach that dynamically adapts the data-stream mining process on the basis of available memory resources. This adaptation is algorithm-independent and enables data-stream mining algorithms to cope with high data rates in the light of finite computational resources. We have also developed lightweight data-stream mining algorithms that incorporate our adaptive mining approach for resource constrained environments.