Foundations of Adaptive Data Stream Mining for Mobile and Embedded Applications

Mining data streams for mobile and embedded applications faces a major problem represented in the high rate of the streaming input with regard to the available computational resources. Adapting the data mining algorithms to the availability of resources is an essential step towards realizing the potential applications in this area. In this paper, we review our Algorithm Output Granularity (AOG) for data stream mining adaptation. The generalization of AOG based on Probably Approximately Correct (PAC) learning model is presented. This generalization is of paramount importance to establish a theoretical framework for adaptation and resource-awareness in data stream mining.