Abstract
Dynamic, data-driven methods are key to enabling the deployment of accurate and efficient data stream mining (DSM) systems, by invoking suitably configured queries in real-time on streams of input data. With the proliferation of technologies for big data analytics, application areas for stream mining are numerous and continually expanding—representative examples include healthcare, climate monitoring, surveillance, and network security. Due to the typically physical separation among data sources and computational resources, it is often necessary to deploy such stream mining systems in a distributed fashion, where local learners have access to disjoint subsets of the data that is to be mined, and forward their intermediate results to ensemble learners that combine the results from the local learners; this is true also in the case of edge computing where computation can take place at the data source. In such a distributed, data stream mining context, DDDAS principles must be incorporated strategically at all levels of the design and implementation process to effectively manage trade-offs among stream mining accuracy, resource requirements, performance, and energy efficiency. This chapter presents methodologies for such DDDAS-integrated, design and implementation of data stream mining systems, referring to these methods as Dataflow- and DDDAS-integrated Adaptive DSM system Design (DDADD), and which combine the methods of dataflow-based signal processing system design with the DDDAS paradigm.