PROBABILISTIC-STATISTICAL ANALYSIS OF THE CORRESPONDENCE BETWEEN SCHEDULED AND REAL-TIME ARRIVAL TIMES OF PUBLIC TRANSPORT BUSES

Read full text Summarise with VIRAT
Authors: Viktor MARTYANOV, Andrzej Czerepicki, M. Kozłowski, Weronika Zaklika
Cited by: 0
References: 24

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Although reliable schedule adherence is pivotal for service quality and passenger planning in urban transit, realized arrivals often deviate from timetables in non Gaussian and time-dependent ways. This study provides a probabilistic–statistical assessment of how actual arrivals accord with schedules, thereby addressing the analytical gap of stop level comparisons across time of day and weekday strata. Using one month of real-time GNSS records matched to static GTFS for Warsaw bus line 112, we analyzed 8.6 million validated stop events. Delay distributions were characterized by robust statistics (median, interquartile range, skewness, kurtosis) and normality diagnostics (Lilliefors, Anderson–Darling, Jarque– –Bera), revealing heavy tails up to ±20 min and invalidating Gaussian assumptions. Non parametric tests (Kruskal–Wallis; Dunn–Šidák) showed statistically significant but practically small differences in median delay across seven time-of-day bands and weekdays (largest shift ≈ 1.2 min; typical deviations within ±1 min). We then clustered stop-level delay profiles (medians by time band) with K-means to uncover four interpretable punctuality archetypes (punctual, evening catch up, pre-schedule variability, and afternoon peak bottleneck), mapped along both travel directions and examined by weekday. We used these profiles to link delay patterns to infrastructure and operations (e.g., initial segments, intersections, recurring congestion) and furnish actionable outputs: targeted buffer allocation, identification of segments for control, and interpretable features to enhance learning-based arrival time prediction. Although demonstrated on a single line, the methodology is general and transferable to other networks with comparable GNSS and GTFS data.

Fields of Study & Subjects

Transportation Planning and Optimization Railway Systems and Energy Efficiency Urban Transport Systems Analysis

Publication details

Journal / source
Transport Problems
Volume
21
Issue
1
Pages
5–18
Publication date
1 March 2026
Language
EN