# Algorithmic Analysis of GTFS-RT vehicle position accuracy

**Authors:** Joshua Wong  
**arXiv:** [2506.06479](https://arxiv.org/abs/2506.06479)  
**Date:** 2025-06-06  
**Summary page:** https://www.summarizepaper.com/en/arxiv-id/2506.06479v1/

## Key points

- Three novel algorithms for calculating geodesic intersections on an ellipsoid
- Analysis of real-time transit data in California to assess vehicle position drift
- Identification of key dataset issues, including missing GTFS FeedMessages and various types of missing data points
- Around 30% of the dataset rendered unusable for analysis due to errors
- Observation of a nightly pattern in the percentage of vehicles within 35 meters of their scheduled route, indicating potential errors like unlinked trips or disabled transponders
- High standard deviation in vehicle distance from the scheduled route possibly caused by errors like stops too far from shape within the GTFS dataset
- Distribution map showing most information originating from San Francisco Bay Area and Los Angeles County
- Alignment of GTFS data with geographical features, despite some inaccuracies compared to OpenStreetMap
- Proposal of practical solutions to improve positional accuracy for both data producers and consumers

## Summary

This paper by Joshua Wong presents three novel algorithms for calculating geodesic intersections on an ellipsoid. These algorithms are applied in a case study analyzing real-time transit data in California to assess vehicle position drift. The analysis reveals that while certain data anomalies can be corrected, large-scale discrepancies persist. The study highlights key issues within the dataset, including missing GTFS FeedMessages and various types of missing data points. These errors render around 30% of the dataset unusable for analysis and raise concerns about the accuracy of the data. Furthermore, the paper discusses a nightly pattern observed in the percentage of vehicles within 35 meters of their scheduled route. This suggests potential errors such as vehicles not being unlinked from trips while in storage or transponders not being disabled during maintenance. The distribution of vehicle distance from the scheduled route also shows a high standard deviation, possibly caused by errors like stops too far from shape within the GTFS dataset. In addition to these findings, a map depicting California's GTFS and GTFS-RT data showcases that most information originates from the San Francisco Bay Area and Los Angeles County. While there may be some inaccuracies when compared to OpenStreetMap, overall the GTFS data aligns well with geographical features. Overall, this comprehensive analysis sheds light on challenges faced in real-time transit data accuracy and proposes practical solutions to improve positional accuracy for both data producers and consumers. By addressing these issues and implementing suggested measures, significant enhancements can be made to enhance the reliability and precision of transit data analysis.