A Concept Drifting Based Clustering Framework for Data Streams

It has attracted extensive interests to discover knowledge from data streams generated in real-time. At present, there are some data streams mining frameworks, providing mining solutions for data streams. This paper proposes an on-demand framework (SRAStream) based on the concept drifting detection. SRAStream allows quick clustering with certain accuracy using only limited resource, enabling the real-time mining of very large data stream with acceptable cost. A concept drifting detecting algorithm is proposed, which employs a quick clustering solution to achieve an accurate detection and then perform the related detecting calculation. Experiments have been conducted based on the UCI datasets. The result suggests that the proposed framework does work well and improve the processing speed greatly in data streams clustering.