Transcript see Jeremy`s poster - Computer Science and Engineering

REU 2007-Automated Mode Detection
Jeremy S. Weinstein, Mentor: Sean Barbeau
Sample Trip Images
Car
Introduction
In the past, census data was collected on
travelers using tedious surveys. Users would often
neglect to fill out these surveys or remember
information incorrectly after their trip. Eliminating
the need for this survey will lead to an increase in
productivity for anyone using this information on
transit. One important question on every survey was
“What was your mode of transportation.”
Point in Buffer Road Map
(Tampa)
Bus
Walk
The Point In Buffer Solution takes a thick line
which represents the location of a bus route and
compares it to all the points of a trip. This
determines if the user is on a bus. This method is
likely to be more effective when combined with a
Critical Point algorithm, but less effective as a large
scale project.
A Neural Network is a program which is
trained by being given input on various trips until it
learns the difference between modes due to subtle
differences such as change in speed, distance
between stops, dwell time, etc.
The Point In Buffer may give the greatest
accuracy during the testing phase, but it is
problematic in the real world. Any bus company that
does not share route information would be seen as a
car. Also, initial testing of this program indicated
that it may require too much processing time to be
viable if made public to multiple users.
The Neural Network, on the other hand, has
become increasingly promising as more test data is
obtained. Thus far it has a 87% accuracy with bus,
car, and walking as the three modes to identify (July
18, 2007). This will likely increase as more data is
obtained for training. However, if a Critical Point
algorithm is applied to the phone, this may go down
drastically.
Methodology
Two possible methods to automatically
determine the mode of transportation are the Point
In Buffer Solution and the Neural Network
Solution. Each method was used in programs which
attempted to detect the mode of a traveler. The
differences between bike, walk, and automobile can
generally be determined with speed. The complexity
lies in the difference between a car or a bus.
Results
Neural Network (Concept)
Acknowledgements
Special thanks to the TRACIT team at CUTR from
USF for their work, expertise, and assistance, the USF
REU program for the opportunity, and Dr. Rafael Perez
for his support.
References
I. Witten and E. Frank "Data Mining: Practical
machine learning tools and techniques", 2nd Edition,
Morgan Kaufmann, San Francisco, 2005.
http://www.cs.waikato.ac.nz/~ml/weka/ July 2007
http://edn.esri.com/ June 2007.
Department of Computer Science & Engineering