A Framework for Trace Clustering and Concept-drift Detection in Event Streams

Concept-drift is a well-known problem that affects data streams where the underlying relations between a recorded tuple x and a system response y change over time [1]. Ignoring concept-drift can lead to a deterioration in the quality of predictive analytics and its capacity to represent the most recent concepts. Nevertheless, implementing a concept-drift adaptation strategy is not a trivial task due to different types of concept-drift and different adaptations in response to them. In this work, we discuss the Concept-Drift in Event Stream Framework (CDESF) that addresses some of these challenges for Trace Clustering (TC) [2] in data streams. Instead of creating an additional level of complexity that isolates the final user from the deep behaviour of the system, our goal is to offer a simple instrument to supervise concept-drift and tracking the evolution of clusters over time.