about me



< September 2010 >
    1 2 3 4
5 6 7 8 91011

past articles »

Click for San Francisco, California Forecast

San Francisco, USA


Traffic Data Analysis

I was doing a traffic data analysis base on the vehicle location data pull from the SF Muni website. This is an extremely interesting project. I pick up a whole lot of new skills while doing this, most notably data analysis and computational geometry. The project also turns out to be a challenging one. A few months (of part time work) has gone by I still haven't nearly achieved my original vision.

Still I think I am getting better at this work. So I'm sharing some interesting data I'm working with. On one bus route I've collected 20,000 location readings on a single day. The goal is to organize them into individual vehicle trip. After a first pass, the data are reduced to 336 trips. But how good is the quality of data? And how good is my trip segregation algorithm? So I plotted the trips' distance against their duration on a chart. The graph quickly helped me to evaluate the quality of the result.

Bus #38 trips characteristic

The first impression is the result looks fairly good. A dense band of points shows that most trip length is about 10 km, and they takes about 40-75 minutes. This seems to match real world experience. At the lower left hand corner is a number of trips that last very short time and cover very little distance. I'll probably treat them as noise and discard them. A curiosity is some trips seem to cover a long distance from 15 to 20 km, much long than the official route. Where did they went?

Closer examination on their track reveal the problem. These are all legitimate eastbound trips. However the bus starts from a depot in the middle of the city. They went all the way to the terminal in the west end. Only from there the real eastbound trip begins. The challenge for me is I'm only interested in the eastbound portion. The trip from the depot to the west end is actually data pollution. I need to find a way to handle these extraneous data.

This still leaves a scatter of points of around 5 km long and last 15 to 35 minutes to inspect. Meanwhile I have difficulty to examine the track because Google Earth is crashing on me all the time. I guess this is a good time to turn away from technical work and to write a blog instead.

2010.09.02 [, ] - comments



blog comments powered by Disqus

past articles »


BBC News


Trump announces in China tariffs (22 Mar 2018)


Trump's top Russia probe lawyer resigns (22 Mar 2018)


Pacific garbage patch 'growing rapidly' (22 Mar 2018)


Kenya payout for mother made to deliver on hospital floor (22 Mar 2018)


Facebook data: What the social media giant knows about you (22 Mar 2018)


Uber self-driving crash: Footage shows moment before impact (22 Mar 2018)


Theresa May warns EU of 'pattern of Russian aggression' (22 Mar 2018)


Ukraine arrests pilot hero Savchenko over 'coup plot' (22 Mar 2018)


Nicolas Sarkozy: French ex-president says funding probe is 'hell' (22 Mar 2018)


EU and six other countries exempted from US metals tariffs (22 Mar 2018)

more »


SF Gate


Bay Area News (7 Jan 2012)


City Insider (11 Feb 2012)


Crime Scene (13 Feb 2012)


C.W Newius Column (10 Jan 2012)


C.W. Nevius Blog (11 Feb 2012)


Education News (10 Jan 2012)


KALW (11 Feb 2012)


Matier and Ross Blog (11 Feb 2012)


YouTube bans firearm sales and how-to videos, prompting backlash (22 Mar 2018)


Uber’s new rival in Australia: an Indian upstart (22 Mar 2018)


Video shows Uber robot car in fatal accident did not try to avoid woman (22 Mar 2018)


‘Blurred Lines’ ruling; medical bankruptcies may be rarer than thought (21 Mar 2018)


Business News Roundup, March 22 (21 Mar 2018)


Uber crash spotlights Arizona as self-driving hub (21 Mar 2018)

more »


Site feed Updated: 2018-Mar-22 12:00