| City: |
Washington , DC , U.S.A
|
| Organization: |
Department of Geography and Geoinformation Science, George Mason University, Fairfax, VA 22030, USA |
| Project Start Date: |
26th June 2018
|
| Project End Date: |
7th August 2018 |
| Reference: |
Robert Truong, Olga Gkountouna *, Dieter Pfoser and Andreas Züfle Department of Geography and GeoInformation Science, George Mason University, Fairfax, VA 22030, USA; rtruong2@masonlive.gmu.edu (R.T.); dpfoser@gmu.edu (D.P.); azufle@gmu.edu (A.Z.)
|
| Problem defination: |
To predict the inflow and outflow of passenger volume at stations of Washington DC Metro. |
| Technical Solution: |
- Principal Component Analysis: This analysis was performed to compress the high dimensional vector compress the data to plot the reduced time series.
- kNN classification analysis: This analysis was performed to classify the clustered model of station and days with respect to time to predict the inflow and outflow of the passenger
- Multi-Layer Perceptron classifier: A similar model used for prediction of the passenger inflow/outflow
|
| Datasets Used: |
Data from the fare card log records was acquired from the Washington Metropolitan Area Transit Authority(WMATA). Each record consists of the station name and the time stamp from when each passenger entered and exited the station.
|
| Outcome: |
- kNN analysis approach showed that the time-series of passenger inflow outflow at stations are highly discriminative that allows to distinguish different stations as well as weekends with extreme high accuracy.
- Whereas the MPL prediction suffers the most on typically weekends when it tends to underestimate the true passengers.
|
| Issues that arose: |
- Anonymous card data. The number of trips per passenger wasn’t known. Only the number of trips is known which makes it difficult to figure out who exited/ entered where.
- Data set of only two months was available so it was difficult to build deep learning algorithm for better prediction.
|
| Status: |
Unknown |
| Entered by: |
Drishya Nair, drishya.nair@mail.utoronto.ca
|