CE-Stream : Evaluation-based technique for stream clustering with constraints

Abstract

Large number of stream clustering techniques have been proposed in recent years. However, these techniques still lack of using background knowledge which are available from domain expert. In this paper, CE-Stream, an incremental method for stream clustering by using background knowledge as constraints is proposed. Instance-level constraint operators are introduced to support evolving characteristics of dynamic constraints i.e. constraint activation, fading and outdating. Constraint operators seamlessly integrate into E-Stream to check active and update constraints and prioritize constraints. Likewise, CE-Stream reduces an excessive splitting during clustering process. Compared to E-Stream, experimental results show that CE-Stream give better clustering performance in terms of both cluster quality and execution-time.