Passenger data complexity in tram stop dwell time modelling

Ahac, Maja; Cvitić, Leonard; Majstorović, Igor; Ahac, Saša; Bašić, Silvio · 2022 · Crossref

DOI: 10.5592/co/cetra.2022.1427

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This study investigates the impact of passenger data complexity on the accuracy of tram stop dwell time prediction models. While previous research suggests that incorporating detailed variables—such as door width, seat count, and specific passenger distribution—improves estimation, such approaches require difficult-to-obtain data regarding passenger distribution inside vehicles and on platforms. The authors aim to determine whether simpler models, relying only on total passenger volumes, can adequately estimate dwell times without significant loss in precision. The research was conducted at an island tram stop in Zagreb, Croatia, selected to minimize interference from car traffic and pedestrians. Data were collected over five working days in October 2020 using video recording devices. The dataset comprised 70 hours of footage, which was processed to identify tram types, arrival/departure times, and passenger counts per door. After filtering for outliers and focusing exclusively on the high-floor TMK301 tram type (which constituted 66% of observations), the final sample included 571 tram stops and 4,762 passengers. The data were split into training (80%) and test (20%) sets for model validation. Three multiple linear regression (MLR) models were developed to predict dwell time based on varying levels of input complexity. The most complex model (MLR-FD) utilized the volume of boarders and alighters through the busiest doors, categorized by flow type (mainly alighting, mainly boarding, or mixed). The intermediate model (MLR-D) used volumes through the busiest doors but ignored flow type. The simplest model (MLR-T) used only the total volume of boarders and alighters per tram. Cross-validation results indicated that simplifying the input data had a minor effect on the model’s goodness of fit. Specifically, the correlation coefficients and coefficients of determination remained similar across all three models. The findings demonstrate that reducing data complexity by using total passenger counts rather than door-specific flows has only a mild effect on model accuracy and precision. The error distribution for the simpler models showed slightly higher standard deviations compared to the complex model, but this discrepancy could be addressed by adding a 3-second operating margin to the simpler models. The study concludes that while these results are specific to the TMK301 tram type and island stop conditions, they suggest that detailed passenger distribution data may not be strictly necessary for adequate dwell time estimation. This simplification facilitates easier model application for timetable creation and traffic planning, though further research is needed to generalize findings to other vehicle types and platform configurations.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• exposure measurement