A Benchmark Dataset for Bus Travel and Dwell Time Prediction

This repository contains the dataset for the paper "A Benchmark Dataset for Bus Travel and Dwell Time Prediction", to be presented at IEEE ITSC 2025. Abstract: The prediction of bus travel and dwell times using machine learning has been extensively studied, resulting in many different approaches. However, due to the absence of standardized benchmarks, the field currently lacks meaningful comparison of model performance. We compile and release a benchmark based on three years of automated vehicle location (AVL) data covering the Dutch public transport network. This includes data excerpts representative of different evaluation scenarios (rural, urban, small and large cities) as well as a methodology for calculating metrics in a reproducible manner. Acknowledgements: The dataset contains "Unofficial archive of travel information Dutch Public Transport" by Adriaan van Natijne, which is licensed under the Creative Commons Attribution 4.0 license. This research was funded by the Ingolstadt public transit authority (Verkehrsverbund Großraum Ingolstadt, VGI) with funds from the German Federal Ministry of Transport (Bundesministerium für Verkehr, BMV) as part of the research project VGI newMIND. Changelog v2 When invidual segments of a trip are invalid, only those segments instead of the entire trip is now dropped. Stop locations are now calculated as the geometric median instead of all records. Fixed bug where ONSTOP events were processed incorrectly. Fixed bug where trips that span LAU boundaries were processed incorectly.