Predictor Picks for Week 15

by mgrex03 on Dec 11, 2008 1:15 AM EST Comment 9 comments

Last week, the Predictor was 10-6 That puts me at 129-79 on the season, which is 62.0%. Still strong.

Here are the picks for Week 15:

Win %	Away	Home	Win %
14.7%	New Orleans	Chicago	85.3%
29.0%	Tampa Bay	Atlanta	71.0%
54.1%	Washington	Cincinnati	45.9%
11.9%	San Francisco	Miami	88.1%
70.0%	Tennessee	Houston	30.0%
15.5%	Detroit	Indianapolis	84.5%
35.8%	Green Bay	Jacksonville	64.2%
32.7%	San Diego	Kansas City	67.3%
10.5%	Buffalo	N.Y. Jets	89.5%
75.2%	Seattle	St. Louis	24.8%
97.8%	Minnesota	Arizona	2.2%
27.5%	Denver	Carolina	72.5%
37.6%	New England	Oakland	62.4%
23.4%	Pittsburgh	Baltimore	77.6%
70.0%	N.Y. Giants	Dallas	30.0%
49.0%	Cleveland	Philadelphia	51.0%

Nideak asked a question last week about where I get the probabilities from, and I didn't see it until today, so I thought I'd answer him here. Basically I've taken every major stat over the past 6 years, created 4 "levels" for each one, and loaded them into a Bayesian Network, created by an algorithm that I don't remember the name of. I then enter in predicted data, which I base off of the prior 7 weeks.

One of the stats loaded into the network is "Winner", so it "knows", based on old data, what has the best chance of winning a game. Once I get all the data in for each of the categories, other than "Winner", I get a probability for each team to win, and that's what I report here.

If interested, I talked a little bit more about it here and here. Feel free to fire more questions away.

0 recs | Comment 9 comments | Share

Story-email Email Printer Print

Comments

Odd(s)

The Vikings away at Arizona more often, run an infinite amount of times, than the Colts beat Detroit at home? That seems bizarre to me, even when given credence to Arizona’s poor historical data.

by Jacob Burch on Dec 11, 2008 11:40 AM EST reply actions 0 recs

See below for an explanation

It isn’t a simulation, but a probabilistic inference.

by mgrex03 on Dec 11, 2008 2:54 PM EST up reply actions 0 recs

I agree witht he Vikes/Cards comment. In addition, Does Oakland’s big margin include Cassell’s departure? Otherwise, I cannot fathom this. I have no problem with the favorites, but the margins seem outsized in general—Chi over NO? Atl over TB? Jax KC Seattle Balt, all seem to be “guaranteed” wins and I just don’t see that. Of course a 1 point win 7 times out of 10 would be a 70% chance….. but a 75/25 split suggests to me that there is litle doubt of the outcome and the winner should dominate. I don’t see Balt dominating Pitt any more than I see Minn annihilating Ariz. And I’d say that the likelihood of us beating Det is greater than Chicago’s of beating NO…. Am I alone in this? Oh well, not a gambler, just an observer.

Bobman

by Bobman on Dec 11, 2008 1:46 PM EST reply actions 0 recs

It is completely based on what has happened over the past 6 years

The difficult part is trying to predict how each team will fare before hand. That is where the majority of the work I need to do in the offseason to improve it will go.

Basically, when it says 70% chance of winning, it means that, looking back at past games, when teams played each other with similar stats, Team A won 70% of the time, and Team B won 30% of the time.

As I said in the other articles, the most influential factor in the whole system is rushing attempts. Rightly or wrongly, the more rushing attempts, the more winning. I understand the argument that winning sometimes leads to rushing attempts (such as a kneel down), and I need to work through those. That’s why the Raiders and Chiefs keep coming up winners against pass-happy offenses. It’s also why it devalues what the Colts and Saints do, as they pass the ball much more than they run.

No computer system is perfect. I’m just trying to make it the best it can be. I’ll be constantly updating during the offseason, and reporting back here. Thanks for the interest.

by mgrex03 on Dec 11, 2008 2:53 PM EST up reply actions 0 recs

Ah, thanks.
I was just perusing the ESPN “experts’” picks, which, as you can imagine, are pretty different, but their win % is about the same as yours. interesting…. Accuscore is at 141-67. The Accuscore commentary for Colts this week:
"The Colts are by far the heaviest favorite of the week. For the Lions to pull off the monumental upset, they would need to 1) Intercept Peyton Manning at least twice; 2) sack him at least 3 times; AND 3) get 100+ rushing yards from Kevin Smith. When all 3 things happen the Lions win 53 percent of the time by an average of 1 point. " That kills me.
The experts are divided onthis, the Accuscore Pitt/Balt percentages there are 52/47 in favor of Balt, which makes more sense to me than the Predictor’s 77/23.
Interesting to watch how your system does.
Isn’t 6 years kind of long in this league to be looking back? Most teams go through a new QB, RB, OC, and HC within 6 years time, no? The 1997 Colts would not help with analysis of the 2000 Colts, for example. We’ve been freakishly consistent, staff and major personnel-wise since then, of course…. Then again, you do need a certain amount of data pionts to get a reasonable conclusion.

Bobman

by Bobman on Dec 11, 2008 7:09 PM EST up reply actions 0 recs

I only predicted results over the past 7 weeks

I found anything beyond that is not significant.

The model’s percentages are based upon 6 years of data. To give a simple example, say Team A has averaged > 33 rush att /game over the past 6 weeks, and Team B is between 23 and 28 rush atts/game. Over the past 6 years, when two teams who faced each other had this situation, Team A won 89% of the time. You then start throwing in other stats, and it keeps changing.

The historical data is a guide. It just tells what stats lead to wins, and predicts it based on a prediction of what the two teams will do.

by mgrex03 on Dec 11, 2008 7:53 PM EST up reply actions 0 recs