Los Angeles Metro Bus Data Analysis Using GPS Trajectory and Schedule Data

Kien Nguyen, Jingyun Yang, Yijun Lin, Jianfa Lin, Yao‐Yi Chiang, Cyrus Shahabi2018

TL;DR

This paper presents a system for analyzing the performance of the Los Angeles Metro bus system using GPS trajectory and schedule data. By processing and integrating these datasets, the authors develop algorithms to measure key performance metrics like travel-time reliability, bus bunching, and on-time performance, visualized through an interactive web application.

What Problem Does This Paper Address?

Public transportation systems in large cities are crucial for alleviating traffic congestion and providing mobility. However, inefficient operations can lead to significant time and cost burdens for commuters. While GPS devices on buses generate vast amounts of trajectory data, effectively storing, managing, processing, and visualizing this data for meaningful performance analysis remains a challenge. This paper tackles the problem of extracting actionable insights from the complex interplay of real-time bus movements and fixed schedules to identify inefficiencies and improve public transit performance in Los Angeles.

Methods

The core of this work lies in a multi-stage data preprocessing pipeline and the subsequent development of analytical metrics.

  • Data Preprocessing:
    • Trip Progress Mapping: Raw GPS records are mapped to stop sequences by identifying the closest stop in a route's trajectory, assuming a non-strictly increasing or decreasing stop sequence for a given route. A parent trip with the longest cumulative distance is used to define the route trajectory.
  • Trajectory Splitting: GPS records are grouped into "runs" representing a bus's journey between start and end stations. This is achieved by calculating a "trend" value based on the change in stop sequence between consecutive GPS points. Consecutive points with the same non-zero trend are considered part of the same trip. Idle points and runs with too few records are removed.
  • Direction Recovery: The direction of each GPS record within a run is recovered based on the identified trend value.
  • Analytical Metrics:
    • Arrival Time Estimation: This crucial step combines GPS data with schedule data to estimate when a bus arrives at each stop. It involves identifying candidate trips from schedules that match the GPS run's time and location, then interpolating arrival times between GPS points and scheduled stops.
  • Travel-Time Reliability (On-Time Performance): Measures the percentage of times a bus arrives within a defined window (1 minute early to 5 minutes late) of its scheduled arrival time.
  • Travel-Time Deviation: Calculates the average difference between scheduled arrival times and estimated arrival times.
  • Bus Bunching: Identifies instances where multiple buses of the same route arrive at a stop too close together, indicating potential operational issues.
  • Waiting Time Estimation: Estimates the average waiting time for passengers, derived from the estimated arrival times.

Datasets

  • GPS Dataset: Collected from the Regional Integration of Intelligent Transportation Systems (RIITS) for the Los Angeles Metro bus system. This dataset includes route ID, bus ID, run ID, timestamp, latitude, and longitude, with approximately one million records per day.
  • Schedule Dataset: Provided by LA Metro in the General Transit Feed Specification (GTFS) format (December 2017 version). This dataset contains information on routes, trips, stops, and scheduled arrival times.

Key Insights and Ideas

  • Bridging Real-time and Scheduled Data: The paper highlights the necessity of integrating dynamic GPS trajectory data with static schedule data to derive meaningful performance metrics. Simply having GPS data isn't enough; it needs to be contextualized by the intended service.
  • The Importance of Preprocessing: The detailed preprocessing steps (Trip Progress Mapping, Trajectory Splitting, Direction Recovery) are critical for cleaning and structuring the raw, noisy GPS data into usable "runs" and "trips" that can be aligned with schedules.
  • Visualizing Performance for Actionability: The development of an interactive web-based application is key. It transforms complex analytical results into easily digestible visualizations, enabling transit agencies to quickly identify problem areas.
  • Multi-faceted Performance Metrics: The paper goes beyond simple on-time performance to include metrics like travel-time deviation and bus bunching, offering a more comprehensive view of operational efficiency and passenger experience.

Key Results

The paper demonstrates the successful implementation of algorithms and a system that can analyze LA Metro bus performance. While specific quantitative results for each metric are not explicitly detailed in summary form, the paper showcases:

  • The ability to process large volumes of GPS and schedule data.
  • The development of algorithms for estimating arrival times, which is a foundational step for all subsequent analyses.
  • The calculation of key performance indicators such as travel-time reliability, travel-time deviation, bus bunching, and waiting time estimation.
  • A functional web-based dashboard that visualizes these metrics, allowing users to explore data at the route, stop, and trip levels. The "Overview Page" shows an average travel-time reliability of 42.05% across all routes.

Limitations and Future Work

  • Limitations:
    • The reliability of the bus direction data in the raw GPS records is noted as poor, necessitating a recovery process.
  • The paper acknowledges that the accuracy of arrival time estimation depends on the density and quality of GPS data and the assumption of constant speed between GPS points.
  • The current system focuses on analysis; real-time prediction capabilities are not the primary focus.
  • Future Work:
    • The authors plan to develop machine learning algorithms for forecasting the presented performance metrics.
  • They intend to incorporate these forecasting algorithms into the current system.

Summary updated

BibTeX
@misc{nguyen2018los,
  author = {Kien Nguyen and Jingyun Yang and Yijun Lin and Jianfa Lin and Yao‐Yi Chiang and Cyrus Shahabi},
  title = {{Los Angeles Metro Bus Data Analysis Using GPS Trajectory and Schedule Data}},
  year = {2018},
}

Loading related content...

Figure