re: DoCho's Bowl Prediction Algorithm: Day 3

Let's not forget this algorithm had Colorado St. as a double digit winner and they needed a horseshoe up their arse to cover.

quote:

---

Did you run this algorithm during the season? It seems bowl season would be a tough time to start an algorithm. Conference games at least a lot of the teams have common opponents. Bowl games are against out of conference teams, which seems like it will result in more error.

---

I started running mine the week of October 12, and tracked the results of all 4 models.

LINK

You can view the results there and use that data to cherry pick various angles as it breaks down the results into a lot of cross tabs.

Unfortunately, since pretty much all bowl games are at neutral sites the home/away breakdowns won't do much good.

quote:

---

It seems bowl season would be a tough time to start an algorithm. Conference games at least a lot of the teams have common opponents. Bowl games are against out of conference teams, which seems like it will result in more error.

---

I agree with all of this.

And those are the reasons why I am doing it

I was very rushed in finishing up my algorithm. There are lots of revisions I can make to it, but I still think it can be successful the way it is. It just may not be as great of a return as Id like

Im working on it right now. I want to finish a new version before the new years eve games.

Your algorithm has a lot of work to do in the 2nd half

quote:

---

Today's game is Boise St vs Oregon State. The lines I see have Oregon State as a 3 point favorite. My algorithm has selected Boise State as the winner. OSU was placed in the cluster expected to lose, usually by double digits, while BSU was placed in the cluster expected to win, usually by double digits. My algorithm gives BSU an 82.4% chance to beat the spread, and also win the game. It gives BSU a 72.5% chance to win the game by double digits

BSU +3. $ in the bank

---

So...

dude you suck

quote:

---

I started running mine the week of October 12, and tracked the results of all 4 models.

LINK

You can view the results there and use that data to cherry pick various angles as it breaks down the results into a lot of cross tabs.

Unfortunately, since pretty much all bowl games are at neutral sites the home/away breakdowns won't do much good.

---

You have the average score as 25.9-30.4. I assume the latter is the home team?

Thanks. This is a pretty clear 33% er.

Im gonna try to do some serious tweaks before the games on the 26.

Here are my tasks:

1. Double the number of data points.
2. Possibly weight them by year for relevance. The game changes significantly year to year
3. Include NCAA SOS as a metric upon which clusters can split.
4. Experiment with the number of clusters to see the resulting ranges and tree purity.
5. Look at correlation between variables in order to prune the tree

Mathematical capping is nearly impossible with bowl season because of motivation and coaching changes. I've done extremely well with my #s during the regular season but generally stay far away from bowl season because of this.

I still cap individual games but don't put nearly as much credence into my comp stats.