Detecting Structural Changes in Urban Road Network Performance: A Kernel-Based Hierarchical Change Point Detection Framework

Abstract

Urban road networks generate increasingly rich long-term traffic datasets, yet their full temporal depth remains underexploited. Most retrospective studies either aggregate data over long periods or perform before–after comparisons tied to pre-identified events, missing any unanticipated change. This paper proposes a general unsupervised framework for automatically detecting all dates at which a structural or relevant change occurred in urban traffic performance. The method combines a set of MFD-based macroscopic indicators and microscopic indicators — capturing the distribution of speed, flow, and density across individual road segments — with a hierarchical kernel-based change point detection algorithm operating at two complementary spatial scales. Applied to seven years of loop detector data in Rotterdam (2016–2023), the framework identifies both major network-wide disruptions — including the COVID-19 mobility collapse — and subtler, spatially localized reorganizations invisible to global aggregation, each assigned an objective impact score. The resulting dated and impact-scored record of network behavioral shifts offers urban practitioners a systematic tool to audit their traffic data, evaluate the effect of past interventions, and identify stable periods suitable for further modelling.