Aussie Government Competition To Predict Commute Times

databuff writes "Last week, Sydney's Minister of Roads, David Borger, launched a $10,000 competition to develop an algorithm that predicts commute times on a major Sydney freeway. The winning algorithm will be used to power predictions on the Sydney live traffic website. The hope is that the predictions will help commuters make informed decisions about when to travel and on what routes, lowering the intensity of peak hour traffic. In its first week, the competition attracted entries from more than 50 teams and 19 countries."

Comments

So if everyone knows the time to avoid

[Aerorae]

won't everyone just follow the same algorithm and end up with traffic peaking at some other time?

Re:So if everyone knows the time to avoid

[twisteddk]

I get the part about lazy.
Apparently the minister of transportation and roads is too. Here in Denmark, we've had nice functionaing algorithms, based on the length of the queue, and the speed at which it moves. And with some nice LED signs that says "traveltime x minutes from here to whereever", and we've had those for.... Well, at least 10-15 years now. A tried and proven system, what's the point of asking for a new algorithm ? Does he want something that makes a GUESS rather than bases itself upon the facts ?

Thats easy

[MichaelSmith]

I used to do this for VicRoads in Melbourne. At peak times a queue builds up at the exit end of the freeway. Normally the city end during the morning peak. Travel time depends entirely on the length of the queue. Maybe RTA NSW should be paying VicRoads for the software I wrote in the 1990s. It can be a discount on the money VicRoads pays for SCATS.

Re:Thats easy

[TapeCutter]

Here's a simple algorithim to find the the end of the evening peak hour in Sydney. Write a bot that looks for the first slashdot story of the day tagged with 'Australia'.

Re:Thats easy

[MichaelSmith]

You know the slashdot web server logs could estimate travel times fairly well by looking at IP addresses for accounts. User stops browsing at work and starts browsing 45 (or in Sydney 90) minutes later from home.

Fail

[hairyfish]

I predict the traffic will be really busy between 7am and 9am, and then again from 4pm to 6pm. Where do I pick up my prize?

Unintended Heisenberg results

[PolygamousRanchKid]

A while back, some folks in France recommended that vacationers should drive after midnight to avoid the August vacation traffic crunch. The result? Tons of folks drove at 03:00 on toll roads . . . however, only one toll booth was open, as usual, for nights. So there were major traffic jams for hours.

So I ask you, Professor John Nash, what should I do out of a Game Theory analysis? If everyone is told, to drive at a certain time, is it better for me, if drive at that time? Prisoners' Dilemma, on the roads.

The economist John Maynard Keynes was once asked, if he thought that the stock market would go up or down. He answered: "It's not important what I think. I invest on what I think, what the most people think."

Re:Unintended Heisenberg results

[Matt_R]

And what with all people using GPS with traffic avoidance build in it.

Does that actually work? I have a GPS with traffic, and live in Sydney. I'm yet to see the traffic avoidance actually work. It tells me there's traffic up ahead, when there isn't. It tells me there's no traffic when there's a crash up ahead.

So yeah, it gets it backwards. The name of the traffic info provider is SUNA Traffic. Guess what SUNA backwards is...

Self defying

[Ignatius]

Predicting commute times and keeping the results secrets vs. predicting commute time and putting them in real time on a public website are two completely different problems. The former ist simply about estimating an output parameter from a set of input parameters so it's basically about approximating a function. The latter contains a nasty feedback loop as the output paramter is in itself an input parameter as it influences the behaviour of the system, so you're basically looking for a fixed point where the publication of the forecast exactly repells as many drivers at it attracts - only these values allow for a stable prognosis. In economics this effect is known as Goodhart's law.

This means that the competition is about a completly different (and much simpler) problem to that which they are eventually trying to solve.

ignatius

IFF the employers even allowed flex time

[dltaylor]

If the employers already allowed flex time, there would be a more even distribution of commuters over several hours, and the problem would smaller to non-existent.

If the employers don't allow flex time (as apparently they don't, as least in useful numbers), it won't matter what information is available to commuters, they'll still have to be arriving at work during peak commuter density periods, and leaving at the corresponding end-of-work-day.

Really want to fix the problem? Maybe you could have an auction every year for employers to buy their preferred work start hours. For 1000 employees, it might encourage them to save a few bucks by buying "off" hours.

Random number generators needed

[syousef]

I predict the traffic will be really busy between 7am and 9am, and then again from 4pm to 6pm.
Where do I pick up my prize?

When you can predict how Sydney's failing infrastructure affects the commute. I was late to work the other day (on late shift) because a water main burst on Paramatta Road and they blocked the road citybound for a couple of hours while they sorted out the mess. Sydney's roads and transport (and other infrastructure obviously) are so poorly planned and maintained that any algorithm that was remotely successful would need several good random number generators.

Re:10 things to do before I die

[ambrosen]

Yes, because obviously that's responsible for holdups, rather than the fact that there's an awful lot of people trying to travel through an awfully small amount of space.