Article · Scientific Journal of Astana IT University · March 30, 2026

DETECTING AND RANKING URBAN BOTTLENECKS FROM PASSIVE SPEED AGGREGATES

Bakbergen Mendaliyev, Didar Yedilkhan, Aidarbek Shalakhmetov

Abstract

Urban traffic congestion remains a persistent problem, yet many cities still lack dense sensor networks, calibrated simulation models, or detailed origin-destination data for operational bottleneck monitoring. This study develops a lightweight framework for detecting and ranking urban bottlenecks using passive probe-based speed aggregates alone. For each road segment, a free-flow benchmark is estimated from high night-time speeds. Hourly median speeds are then converted into travel times, and cumulative delay is normalized by segment length and mildly regularized to reduce instability on short urban links. The framework is applied to a 20-day probe-data sample for Astana containing 6.34 million link-hour observations across 22,333 segments. After quality checks and coverage filtering, 8,634 segments remain in the final analysis set. The main results are as follows: The estimated free-flow benchmark aligns closely with posted speed limits and remains stable under alternative percentile and night-window definitions. Congestion is strongly concentrated: 1,336 segments account for about 50% of the total delay, while 3,868 segments account for 80%, indicating a pronounced Pareto-type structure. The ranking remains robust after excluding days affected by major external disturbances, which suggests that the main bottleneck pattern is not driven by a small number of atypical days. Spatial diagnostics reveal significant positive autocorrelation in congestion severity, and the clustering pattern becomes stronger when road connectivity is represented with a network-based weight matrix rather than a purely geometric nearest-neighbour specification. Local cluster analysis identifies corridor cores of severe delay together with adjacent transition links, showing that the most critical bottlenecks are spatially connected rather than randomly scattered. Clustering of normalized 24-hour delay profiles reveals three evening-oriented regimes that differ mainly in congestion intensity. Taken together, these findings show that routinely collected passive probe data can recover meaningful and operationally useful congestion structure even when a city lacks dense fixed-sensor coverage or a calibrated simulation model. The proposed workflow is transparent, reproducible, and suitable for corridor prioritization, before-and-after evaluation, and future digital-twin-based traffic management.

Transportation Planning and OptimizationTraffic Prediction and Management TechniquesTraffic control and managementtraffic congestioncongestion delayPareto distribution

DETECTING AND RANKING URBAN BOTTLENECKS FROM PASSIVE SPEED AGGREGATES

Abstract

Related publications