Revealing Trace Variant Shift via Multi-dimensional Profiling and Community-Aware Graph Modeling

Tracking the evolution of trace variants over time is a challenging task in process mining, particularly in dynamic environments where process executions change independently. Existing approaches often fall short in detecting fine-grained, temporal changes at the variant level. In this paper, we present a novel technique that integrates multi-dimensional profiling with graph-based analysis to monitor variant-level shift. Each trace is represented through a comprehensive profile that combines control-flow and time-related information, including features extracted via discrete wavelet transform. These profiles are embedded into a similarity space and linked using a k-nearest neighbor graph to capture local structural patterns. By applying community detection, we identify clusters corresponding to distinct trace variants and analyze their evolution over time. This approach enables the detection of subtle, localized variant shifts, facilitating faster and more informed decision-making within organizations. Experiments on real-world event logs demonstrate the effectiveness of our method in revealing significant behavioral shifts from a variant-centric perspective.