Principal Investigator

Aaron Steinfeld

Status

Completed

Start Date

Jan. 1, 2016

End Date

May 31, 2017

Project Type

Research Advanced

Grant Program

MAP-21 TSET National (2013 - 2018)

Grant Cycle

2016 TSET UTC

Visibility

Public

Abstract

Knowledge about inefficient trips resulting from inconvenient and missed transit connections is sparse and incomplete. Some of this information can be deduced from smartcard fare data by looking at points of payment across each piece of the trip. However, this only provides limited insight on the intended trip of the rider due to the lack of origin-destination (O-D) ground truth. Likewise, it is not possible to disambiguate intentional gaps between trips (e.g., running an errand) from missing a connection. Missed connections and long waits can be inconvenient, suppress optional travel, and expose riders to inclement weather and crime. Poor connections and the resulting exposure to adverse conditions can also motivate riders who have the option of driving or using paratransit to elect a more expensive and inefficient travel mode.

The Tiramisu Transit app collects data on where users are, which bus stops they seek data about, and which trips they take (when tracing). We have used this data to do preliminary O-D analyses and initial identification of connecting trips. In this project, we seek to use Tiramisu data to create a better understanding of where and when poor connections are occurring in the Pittsburgh region. The resulting information will be provided to Port Authority and local municipalities for use during planning and identification of locations where bus shelters and other infrastructure improvements may facilitate more safe and comfortable waits. We will use generalizable methods to ensure broader value to both science and practitioners.
    

Description

Motivation

Knowledge about inefficient trips resulting from inconvenient and missed transit connections is sparse and incomplete. Some of this information can be deduced from smartcard fare data by looking at points of payment across each piece of the trip. However, this only provides limited insight on the intended trip of the rider due to the lack of origin-destination (O-D) ground truth. Riders may pay at one end of the trip but not reveal the other end. This makes it difficult to determine if a rider used a common stop for the connection or a wider gap in order to run errand or another pedestrian activity before connecting. Likewise, it is difficult to determine the initial origin or destination of the overall trip, thus making it very hard to assess demand for the more direct routes. It is possible to make some initial guesses based on historical patterns, but our analyses using ground truth O-D have shown performance on historical estimates to be rather poor.

Missed connections and long waits can be inconvenient, suppress optional travel, and expose riders to inclement weather and crime. Riders with the option of driving may abandon transit (or never chose it) due to regular poor connections. Poor connections and the resulting exposure to adverse conditions can also motivate riders who have the option of using paratransit to elect the more expensive travel mode. 

Unfortunately, it is also hard to disambiguate intentional and unintentional connection gaps. As implied above, it is currently not possible to easily disambiguate intentional gaps between trips from missing a connection with traditional data sources. For example, a gap in a connection could be due to either missing the second trip or due stopping to get coffee at a nearby shop. Therefore, it would be useful to have data mining tools for removing intentional gaps from certain analyses. Similarly, knowledge of which connections that have a high incidence of intentional gaps could be used to identify features which are valuable to riders when connecting.

In this project, we seek to use data from the Tiramisu Transit app to create a better understanding of where and when poor connections are occurring in the Pittsburgh region. The resulting information will be provided to Port Authority and local municipalities for use during planning and identification of locations where bus shelters and other infrastructure improvements may facilitate more safe and comfortable waits. We will use generalizable methods to ensure broader value to both science and practitioners.

Approach

The Tiramisu Transit app collects data on where users are, which bus stops they seek data about, and which trips they take (when tracing). We log time, date, location, and which data is requested per user whenever the app is opened or new data is requested. We regularly obtain ground truth O-D data whenever our riders trace their trips.

We have used this data to do preliminary O-D analyses and initial identification of connecting trips. Under a related Metro21 effort we are trying to infer O-D data from the non-tracing data. For example, if we see a user open the app at a specific bus stop, can we infer their intended destination? By looking at this data for the riders who trace, we can use the trace data as the ground truth. In addition, we can look at trips that share nearby O-Ds in close temporal proximity to find connecting trips. 

In the proposed effort we will use trace data, and any effective O-D inference methods from the Metro21 project, to find trips with connections in them. We can then look at the spatial and temporal gaps during connections to understand where and when poor connections are occurring. Since we also capture all the AVL data for the system, we can conduct retrospective analyses on whether the riders just missed a connection or fall within planned headways.

We can also identify intentional gaps by examining lookup location and timing data during the gap. As mentioned, we log location and time whenever riders request data in the app. This allows us to see if they are at a connecting stop or elsewhere. Riders will often check the arrival times for their second trip while running errands or getting coffee so they know when to head to the stop. Likewise, riders will often open the app at the connecting stop to see when the next one will arrive.

Tasks

1.	We will first refine our preliminary data mining methods for automatically finding connections within the Tiramisu data stream.

2.	Next, we will develop metrics for characterizing connection gaps. These will include traditional measures from the literature, as well as ones specific to data that contains O-D and arrival time data requests. For example, repeatedly requesting arrival times can suggest frustration and disappointment. The new version of Tiramisu also includes the option to attach notes to stops, trips, and routes. We will look for complaints about missed connections and long wait times.

3.	We will then develop methods for disambiguating intentional from unintentional gaps. Our focus will be on long gaps since these are more damaging and harder to disambiguate.

4.	Finally, we will develop visualizations for use by Port Authority planners and municipal stakeholders. Depending on ease of integration, this will either be independent or a data stream that can be visualized within Sean Qian’s mobility analytics center effort.

We will schedule meetings with Port Authority representatives during Tasks 2 and 4. This will allow us to formulate useful metrics (Task 2) and then brief agency representatives on our findings (Task 4). We will publish papers in venues as appropriate.

Team

This effort will be led by Aaron Steinfeld, the PI and Co-Director of the Rehabilitation Engineering Research Center on Accessible Public Transportation (RERC-APT). The RERC-APT is the primary funding source for Tiramisu research and development, thus allowing the team to obtain synergistic support for this project, leading to significant cost savings. The RERC-APT is responsible for all data gathering, data storage, servers, and associated technical support. Steinfeld’s funded effort will be dedicated to management of this project and stakeholder contact. Tomasic and Zimmerman will participate through their regular RERC-APT roles.

The bulk of the data analysis and visualization efforts will be done by one or more undergraduates. Several undergrads already work for our team on Tiramisu topics. One is Sunny Zheng, who we think would be a good candidate for this specific project. It is unclear if he will stay in Pittsburgh for the summer since he has taken internships out of town for the past two years. We sometimes hire hourly MS students ($15/hr) for similar efforts. 

We will partner with Tiramisu Transit LLC, a RERC-APT spinout, to obtain current data from the app. We already have an established relationship with the Port Authority of Allegheny County (PAAC) that predates Traffic21 and the UTC as a result of the RERC-APT. Prior meetings have included discussions with PAAC about how Tiramisu data can be used to identify potential demand for better service along popular connection trips. Since many connections occur within the City of Pittsburgh, we may also discover new opportunities for unmet bus shelter locations, amenities, and infrastructure. The City manages most of the bus shelters within the city limits due to the franchise bid. 

Timeline

We will start Task 1 during the Winter semester (Jan-May). Task 2 may start during this period, depending on part-time student availability and progress. Concentrated effort on Tasks 2 and 3 will be done by a full-time intern over the summer. We expect Task 4 to be completed by a part-time student during the Fall semester.

Strategic Description / RD&T

Deployment Plan

Under Task 4, we will develop visualizations for use by Port Authority planners and municipal stakeholders. Depending on ease of integration, this will either be independent or a data stream that can be visualized within Sean Qian’s mobility analytics center effort.

Expected Outcomes/Impacts

Under Task 2 we hope to develop metrics that are useful descriptors of connection quality. Task 3 will generate new methods for disambiguating intentional and unintentional gaps. Task 4 will demonstrate new views on where and when poor connections are occurring in Pittsburgh.

Expected Outputs

TRID

Individuals Involved

Email	Name	Affiliation	Role	Position
steinfeld@cmu.edu	Steinfeld, Aaron	Robotics Institute	PI	Faculty - Research/Systems
tomasic@cs.cmu.edu	Tomasic, Anthony	LTI	Co-PI	Faculty - Research/Systems
johnz@cs.cmu.edu	Zimmerman, John	HCII	Co-PI	Faculty - Research/Systems

Budget

Amount of UTC Funds Awarded

$50136.00

Total Project Budget (from all funding sources)

$50136.00

Documents

Type	Name	Uploaded
Presentation	Identifying Commonly Used and Potentially Unsafe Transit Connections with Crowdsourcing	April 24, 2017, 8:54 a.m.
Progress Report	33_Progress_Report_2016-12-31	Oct. 5, 2017, 10:03 a.m.
Final Report	33_-_modeling_transit_patterns_via_mobile_app_logs.pdf	June 21, 2018, 8:38 a.m.
Publication	Identifying commonly used and potentially unsafe transit transfers with crowdsourcing	Oct. 24, 2020, 5:56 p.m.
Publication	Accessible Transportation Technologies Research Initiative	Oct. 24, 2020, 7:01 p.m.
Publication	A Long-Term Evaluation of Adaptive Interface Design for Mobile Transit Information	Oct. 24, 2020, 7:05 p.m.
Publication	" You are asking me to pay for my legs" Exploring the Experiences, Perceptions, and Aspirations of Informal Public Transportation Users in Kampala and Kigali.	Oct. 24, 2020, 7:11 p.m.
Publication	Accessible Transportation Technologies Research Initiative (ATTRI): Assessment of Relevant Research	Oct. 24, 2020, 7:13 p.m.

Match Sources

No match sources!

Partners

No partners!