Developing an Origin-Destination-Transfer (ODX) Model

Reading Time: 5 minutes

Pioneer Valley Transit Authority (PVTA)

Introduction

The Pioneer Valley Transit Authority (PVTA) is the second-largest transit agency in Massachusetts. PVTA provides public transportation services across a 600 sq. mi. area covering a population of about 600,000 in Hampden and Hampshire counties. In fiscal year 2019, PVTA provided over 10 million fixed-route rides and over 250,000 paratransit trips.

The map shows PVTA’s 34 fixed bus routes, with hubs in Springfield, Holyoke, Northampton, and Amherst. PVTA serves 24 member communities in Hampden and Hampshire counties, who all contribute a portion of PVTA’s operating cost based on the number of fixed route miles and transit passengers served. In the south, Hampden County is home to a major urban and economic center in Springfield, driven by its education, health and business services. To the north in Hampshire County, the PVTA serves a large student population in Amherst and Northampton, which includes the University of Massachusetts-Amherst, the Commonwealth’s flagship campus, and other Five College area institutions. PVTA is critical to the functioning of the region’s many academic institutions. Approximately 50% of all PVTA trips during the academic year are made by students and staff of the region’s schools, colleges and universities. The remaining trips are predominantly for work, medical and shopping purposes.  

In 2021, PVTA was awarded $617,000 through the Accelerating Innovative Mobility (AIM) grant from the Federal Transit Administration. With this funding, PVTA has focused on developing an origin-destination-transfer (ODX) model that will make use of the latest advances in machine learning to predict passenger trajectories and infer transfer and destination locations of bus transit trips. The project also features the integration of PVTA’s app suite to enable access to real-time transit information and fare payment using TransitApp, and the installation of on-bus mobile ticket validators to enhance fare payment and data collection through mobile technology.

The validator software is equipped to process fare payment from not only smartphones or smart cards, but also other sources such as bank cards and wearable devices that PVTA could consider in the future.

PVTA hopes that the upgrades from this project would make using transit easier and deliver a more seamless travel experience for riders. In addition, the data from the validators would be used to improve the ODX model, and consequently provide PVTA with more detailed information about travel patterns on the existing transit network. The insights will facilitate more effective and comprehensive transit planning

To develop the model and reach the project’s goal, PVTA is partnering with the Pioneer Valley Planning Commission (PVPC), and the Networks for Accessibility, Resilience and Sustainability Laboratory (NARS Lab) at the University of Massachusetts, Amherst’s Civil and Environmental Engineering department.

Since PVTA’s riders are the end users of the app and fare payment upgrades, PVTA will also solicit feedback and satisfaction through a customer survey during the project’s implementation. PVTA would be the end user of the ODX model to ensure adequate services are provided to ultimately better serve its customers.

Building an ODX Model with a Limited Data Warehouse

The ODX model is a program developed in Python that takes in data from PVTA’s existing data warehouse—including data from automatic passenger counters (APCs), global positioning systems (GPS), general transit feed specification (GTFS), and automatic fare collectors (AFCs). While these data sources provide information on where, when, and how many customers are boarding and alighting at PVTA bus stops, they do not provide segment-level information about either the final destination or transfer points of trips. PVTA’s AIM project proposed to develop an ODX model to infer destination and transfer locations for origin trips using information from the mobile ticketing data. 

For example, in the map above, the left image processed from the APC shows the boarding volumes at each bus stop but doesn’t provide the trip links. The right image processed from the mobile tickets and ODX model includes the trip links, highlighting that on this route 36% of trips originate and end/transfer on the northern half of the route, 55% on the southern half, and 9% of trips are between the northern and southern halves. The additional information about how riders are traveling on the route helps PVTA plan improvements that would better serve passengers.

How do we validate the ODX model if we don’t know true origin-destination-transfer information?

One of the project goals was to develop an ODX model to infer trip destinations and transfers. However, to ensure the model provided accurate, representative, and robust outputs, it needed to be validated against real data. This was a challenge because the information PVTA was trying to infer from the ODX model was the same information needed to validate the model.

The study team decided to use customer survey data to elicit true origin, destinations, and transfer locations from a representative sample of customers, which would form the basis for the validation dataset. The behavioral insight gleaned from the customer survey data was also used to calibrate the model.

From this approach, PVTA found that multidisciplinary research can create new insights about transit questions and, in particular, how qualitative and behavioral data can be combined to enhance quantitative research. The team conducted a survey of customers who provided details about their trip trajectories, which in turn provided the core of the dataset used to validate the ODX model. The success of this approach reinforces the relevance of conducting integrative research, using qualitative and behavioral information to improve quantitative and predictive models. Below are screenshots from the ODX dashboard. The first image shows a heat map of where passengers begin their transit trips with the color representing the number of daily trips. The second image with daily trip “desire lines” shows where customers are traveling to and from on the existing network. The blue lines show the most popular origin-destination pairs and highlight the hub and spoke nature of the system. The heat maps and desire lines reveal areas of strong travel demand and connections in the existing network, as well as gaps or inefficiencies in the network, informing PVTA about areas to prioritize efforts to improve transportation services.

Conclusion

The next step in PVTA’s AIM project will be to begin evaluating the ODX model’s accuracy. That process involves estimating the proportion of flows whose magnitudes are predicted with 5% precision using the ODX model, with data sampled from several synthetically generated rider datasets stratified by the percentage of total trips provided as simulated data. 

The ODX model and dashboard deployment are expected to be finalized by June 2023.

For more information about PVTA’s AIM project, please contact Tolu Oke at tboke@pvta.com