[ProgSoc] Predict Sydney traffic?

marauder marauder at marauder.net.au
Fri Dec 3 10:23:22 EST 2010

Le 03/12/2010 à 5:02 AM, John Elliot a écrit :

> If ARIMA won't win any prizes in the competition, what would?

Let me explain: if a professional got handed this problem at work, they'd use ARIMA or something in that family, check that the forecasts were good enough, and move onto the next problem.  But the competition will likely be won by someone from the 'crazy Russian mathematician' or 'student with infinite time' schools of data analysis, by applying something special-purpose that nobody else has ever heard of or by chimping at the problem randomly until something sticks fantastically well, respectively.  

> Implementing anything based on that explanation alone is beyond me. 

And here I was, thinking you wanted 'broad strokes'!  If you want to implement, implement ARIMA.  See [http://www.duke.edu/~rnau/411arim.htm], which teaches you how to do ARIMA in a spreadsheet.  Each formula is a different way to estimate the value at t given t-1 and sometimes t-2, each greek letter is a value you can play with (and one day estimate properly!) to see how it affects the accuracy of the estimates.  To make forecasts, you just insert blank rows into your data series and the formula will give estimates.  I would also do graphs, they always help me.


More information about the Progsoc mailing list