university of washington handling bias in large urban...

1
University of Washington Handling Bias in Large Urban Datasets Using Smart Fare Card Data for Transit Planning Mark Hallenbeck, Mayuree Binjolkar, Daniel Dylewsky , Andrew Ju, Wenonah Zhang, Michael Wolf STATISTICAL BIAS ANALYSIS Sponsoring Agencies Origin-Destination Data from Orca is Biased Against Short Duration Trips Chose Semi- Supervised Based Machine Learning Approach for Classification of Types of Transfers Real Transfer Financial Transfer Transfer Classification Results by Stop Marker Size=Transfer Count Not all transit users are ORCA users Need to account for cash-paying customers for business planning Compare limited passenger count data to ORCA data ORCA can combine Two Trips into one Trip A + Trip B = long trip with transfer Research Task: To determine if a transfer was real or financial If transfer is financial then create new origin – destination pairs MORE FALSE TRANSFERS MORE REAL TRANSFERS ORCA Underestimates Total ORCA = Total ORCA Overestimates Total Outcome of Transfer Classification Number and Type of Transfer Activity Example of ORCA Financial Transfers APC (Automatic Passenger Count) and ORCA Factors by Traffic Analysis Zones Future Work Rebuilding origin-destination data files using transfer analysis results Deploy Hidden Markov Models for statistical bias analysis

Upload: others

Post on 26-Jul-2020

3 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: University of Washington Handling Bias in Large Urban Datasetsescience.washington.edu/wp-content/uploads/2018/01/Final-Poster-ORCA-2017.pdfUniversity of Washington Handling Bias in

University of Washington

Handling Bias in Large Urban DatasetsUsing Smart Fare Card Data for Transit Planning

Mark Hallenbeck, Mayuree Binjolkar, Daniel Dylewsky , Andrew Ju, Wenonah Zhang, Michael Wolf

STATISTICAL BIAS ANALYSIS

Sponsoring Agencies

Origin-Destination Data from Orca is Biased Against Short Duration Trips

Chose Semi- Supervised Based Machine Learning Approach for Classification of

Types of Transfers

RealTransferFinancialTransfer

Transfer Classification Results by Stop

MarkerSize=TransferCount

• Not all transit users are ORCA users• Need to account for cash-paying

customers for business planning • Compare limited passenger count data

to ORCA data

ORCA can combine Two Trips into oneTrip A + Trip B = long trip with transfer

Research Task:•To determine if a transfer was real or financial•If transfer is financial then create new origin –destination pairs

MOREFALSETRANSFERS

MOREREALTRANSFERS

ORCAUnderestimatesTotal

ORCA=Total

ORCAOverestimatesTotal

Outcome of Transfer Classification

Number and Type of Transfer Activity

Example of ORCA Financial Transfers

APC (Automatic Passenger Count) and ORCA Factors by Traffic Analysis Zones

Future Work• Rebuilding origin-destination data files

using transfer analysis results• Deploy Hidden Markov Models for

statistical bias analysis