Graph-Based Alignment of Heterogeneous Tracking Events - A Process Mining and Structural Time Warping Approach to Canonicalizing Shipment Lifecycles

Abstract

In last-mile logistics, tracking shipment progress is obstructed by data heterogeneity. Different companies and services report tracking events using disparate naming conventions, frequencies, and operational sequences. This inconsistency prevents efficient and reliable multi-carrier analysis of shipment lifecycles and the detection of deviant shipment behaviors. To bridge this gap, this thesis introduces a unified pipeline to align and canonicalize heterogeneous tracking events into a standardized shipment lifecycle. The proposed methodology integrates process mining with structural graph theory. First, an optional pre-processing layer utilizes the HeuristicsMiner algorithm to isolate the main underlying process from noise, outliers, and low-frequency traces. Next, the pipeline usesWeisfeiler-Lehman (WL) subtree kernels to project the resulting directed networks into a vector space. This representation enables K-means clustering to group structurally similar lifecycles. These clusters are then used to extract representative reference models known as "golden paths." Finally, cross-carrier nodeto- node event mapping is achieved by applying Structural Dynamic Time Warping (SDTW), followed by a customized heuristic layer. Empirical validation shows that pre-processing through process mining substantially strengthens cluster stability. Furthermore, alignment fidelity was confirmed utilizing a specialized validation framework, demonstrating that the pipeline successfully bypasses superficial string-naming variations to resolve the true logistical intent of ambiguous cross-carrier event sequences. Keywords: