Computer System follow along

[Indecent]

I'm not going to give too many details about how the system is constructed or any of the stats used to make the predictions. I will, however, give a high level overview of how the system works and what I hope to achieve with my technique.

In general, the system has no concept of teams, weeks, years, scores, downs, or even football. It is learning to recognize patterns in numbers and is only as good as the numbers I feed it. I'm using a commonly used artificial machine learning technique that is fed stats and learns to predict the outcome.

After training the prediction systems, I used another AI technique to classify games according to similarity. Once again, the classification system has no knowledge of football or even what the numbers it is grouping represent.

The goal of grouping is to try to find games that are similar, and use the predictors who have had success with this type of game.

The following are some of the predictor results and grouping information from my system, with some commentary from me on each game about what I was looking for and what I like about the grouping.

To read the numbers, just pay attention to the labels at the top. Here is a translation:

Acc = Overall accuracy
HPred = Home team predicted %
H Acc = Home team accuracy , with similar values for the away side.
SP Acc = spread accuracy
Fv Pred = favorite predict %
Fv Acc = favorite accuracy , with similar values for the away side following.
Av Points missed = the average number of points the predictor was off for games in that group.

I also bolded the relevant percentages for each predicter, where away percentages are bold when the predictor selects the away side, and the opposite for home. The fav and dog values are bolded similarly.

I know very little about football and rely entirely on this system for placing my bets, so please feel free to point out any games that strike you as off due to injuries, suspensions, or your own reasoning.

Seattles -3 vs Arizona
I don't like how far off the predictor is for scores in this group (15.7), but it shows a good mix of home away picks, low amount of favorites selected but is very good when it does pick the favorite.
Predicter 1: 11

Acc      HPred   H Acc   AwPred  Aw Acc  SprAcc  FvPred  FavAcc  DogPred DogAcc  Pts Missed avg
0.6268  [B]0.614   0.6687[/B]  0.386   0.5602  0.5148  [B]0.3493  0.5328[/B]  0.6045  0.5413  
Games in this group: 23
0.4783  [B]0.5217  0.6667[/B]  0.4783  0.2727  0.5652  [B]0.2174  0.6  [/B]   0.7826  0.5556  15.6957

Det +13.5 vs Green Bay
I don't like this predictor favors the home team, but I do like how much its dog prediction accuracy increases.
Predicter 1: 10

Acc      HPred   H Acc   AwPred  Aw Acc  Sp Acc  FvPred  FavAcc  DogPred DogAcc  Pts Missed avg
0.6299  [B]0.735   0.643 [/B]  0.265   0.5934  0.5172  0.2877  0.5455  [B]0.6688  0.5373[/B]  
Games in this group: 22
0.7727  [B]1       0.7727 [/B] 0       0       0.5455  0.2727  0.3333  [B]0.7273  0.625[/B]   12.8182

Min -3 vs Balt
What I liked here is both predictors have prediction stats and results with a solid record. Also, the 2 of the three best predictors agreed (both shown below), with both predicting Min covers. Neither side predicts the favorite often, but are very successful when they do.
1: 5
Predicter 1: 11

Acc      HPred   H Acc   AwPred  Aw Acc  Sp Acc  FvPred  FavAcc  DogPred DogAcc  Pts Missed avg
0.6268  [B]0.614   0.6687[/B]  0.386   0.5602  0.5148  [B]0.3493  0.5328 [/B] 0.6045  0.5413  
Games in this group: 38
0.5263  [B]0.8947  0.5588[/B]  0.1053  0.25    0.5     [B]0.3158  0.5833 [/B] 0.5263  0.6     10.6579

Predicter 2: 5

0.628   [B]0.6166  0.6689[/B]  0.3834  0.5621  0.5113 [B] 0.2733  0.5535[/B]  0.6699  0.5361  
Games in this group: 38
0.5789 [B] 0.8947  0.5882[/B]  0.1053  0.5     0.6579  [B]0.3158  0.75[/B]    0.6316  0.6667  10.4737

Cinc -4.5 vs Houston
I like the big difference between the overall prediction tendencies and group prediction tendencies. It is balanced at predicting home/away for all games in my database, but almost always predicts the home team in this grouping and does so with comparable accuracy. Predicter 1 predicts the fav covering much more often with 75% success rate over these games. Predicter 2 shows a smaller increase in predicting the fav in the group but shows a good profit over the 24 games.
Predicter 1: 17

Acc      HPred   H Acc   AwPred  Aw Acc  Sp Acc  FvPred  FavAcc  DogPred DogAcc  Pts Missed avg
0.6227  [B]0.5342  0.69[/B]    0.4658  0.5455  0.4988[B]  0.3546  0.5137[/B]  0.5974  0.5301   
Games in this group: 13
0.6923  [B]1       0.6923[/B]  0       0       0.6154 [B] 0.6154  0.75[/B]    0.3077  0.5     10.3077

Predicter 2: 22

0.6374  [B]0.6518  0.6671[/B]  0.3482  0.582   0.5059  [B]0.4151  0.522[/B]   0.5316  0.5422 

Games in this group: 24
0.7083  [B]0.9583  0.6957[/B]  0.0417  1       0.5833  [B]0.625   0.6 [/B]    0.3333  0.625   11.6667

Atl -3 vs Bears
I included three groupings in this one but all found the same predictor to be the best in the group. Once again, I like the big difference in prediction home/away between the overall and the groupings. I also like the difference between favorite in the first two groups, with the predictor showing more balance between fav and underdog. I think its interesting to point out that the last grouping tended to predict the underdog more in the grouping but still passed on taking Chicago.
Predicter 1: 11

Acc      HPred   H Acc   AwPred  Aw Acc  Sp Acc  FvPred  FavAcc  DogPred DogAcc  Pts Missed avg
0.628   [B]0.6166  0.6689[/B]  0.3834  0.5621  0.5113  [B]0.2733  0.5535[/B]  0.6699  0.5361  
Games in this group: 29
0.7586  [B]1       0.7586[/B]  0       0       0.6071  [B]0.5172  0.642[/B]9  0.4483  0.6154  10.4483 
Games in this group: 23
0.4615  [B]1       0.4615[/B]  0       0       0.3846  [B]0.4615  0.5 [/B]    0.3846  0.4     11.8462 
Games in this group: 19
0.8947  [B]1       0.8947[/B]  0       0       0.4211  [B]0.2105  0.5[/B]     0.7895  0.4     13.3684

Kansas city chiefs +6.5 vs Wash Skins
The results from all the groupings were pretty split, with very few being able to reach a consensus among the three best networks in the group. I like the tendency of predicting favorites dropping in the groupings, although the accuracy of the first predictor against the spread concerns me. Despite the score of the third predictor being very close the spread, I like its accuracy predicting the dog and how low its missed points over the 25 games in the group. and it is lights out against the spread in this group (70%). Despite the split, I like this bet and would not be surprised if the chiefs won outright.
Predicter 1: -7

Acc      HPred   H Acc   AwPred  Aw Acc  Sp Acc  FvPred  FavAcc  DogPred DogAcc  Pts Missed avg
0.6268  0.614   0.6687  [B]0.386   0.5602 [/B] 0.5148  0.3493  0.5328  [B]0.6045  0.5413  [/B]
Games in this group: 14
0.6429  0       0       [B]1       0.6429[/B]  0.2143  0.4286  0       [B]0.5714  0.375[/B]   12.7857

Predicter 2: -5

0.5898  0.6488  0.6311 [B] 0.3512  0.5135[/B]  0.5121  0.2862  0.5271  [B]0.6794  0.5308 [/B] 
Games in this group: 14
0.4286  0.3571  0.2     [B]0.6429  0.5556[/B]  0.5     0.1429  0       [B]0.8571  0.5833  [/B]13.2143

Predicter 3: 4

0.628   0.7346  0.6418  [B]0.2654  0.5897 [/B] 0.5176  [B]0.2964  0.5415[/B]  0.6616  0.5386  
Games in this group: 25
0.64    0.88    0.5909  [B]0.12    1[/B]       0.7083  [B]0.28    0.7143[/B]  0.64    0.8     8.12

Denver +3 vs San Diego
For the first predictor, I really like the big increase in the away team prediction in both grouping schemes. The spread accuracy of this predictor in the first grouping concerns me a little, but looking at this predictor in the other groupings it appears overall it is much more accurate at predicting underdogs than the first grouping shows and appears to be well suited for this game.
Predicter 1: -6

Acc      HPred   H Acc   AwPred  Aw Acc  Sp Acc  FvPred  FavAcc  DogPred DogAcc  Pts Missed avg
0.6268  0.614   0.6687  [B]0.386   0.5602 [/B] 0.5148  0.3493  0.5328  [B]0.6045  0.5413 [/B] 
Games in this group: 21
0.7143  0.0952  1       [B]0.9048  0.6842[/B]  0.45    0.619   0.5833 [B] 0.3333  0.2857[/B]  8.2381  
Games in this group: 19
0.6316  0.2105  0.25   [B] 0.7895  0.7333 [/B] 0.5294  0.2632  0.25    [B]0.7368  0.6154 [/B] 7.3158

Predicter 2: -1

0.6295  0.7868  0.6333  [B]0.2132  0.6152[/B]  0.5145  0.2091  0.5574 [B] 0.7452  0.5332 [/B] 
Games in this group: 19
0.5789  0.4737  0.4444  [B]0.5263  0.7  [/B]   0.6111  0.1053  1      [B] 0.8947  0.5625  [/B]9.2632

NYJ -10 vs Buffalo
I really like this grouping scheme, it is the most reliable in my back testing and seems to do a really good job predicting games with big Spreads. I I like how accurate both predictors are at selecting the favorite to cover, especially considering how infrequently both predict it. For this game, all predictors but 1 have NYJ winning by more than 12.
Predicter 1: 12

Acc      HPred   H Acc   AwPred  Aw Acc  Sp Acc  FvPred  FavAcc  DogPred DogAcc  Pts Missed avg
0.6276  [B]0.6405  0.6623[/B]  0.3595  0.5657  0.5047  [B]0.3539  0.527[/B]   0.5913  0.5373  
Games in this group: 34
0.7647  [B]0.9412  0.7812 [/B] 0.0588  0.5     0.5588  [B]0.3235  0.7273 [/B] 0.5882  0.55    8.9412

Predicter 2: 18

0.6181 [B] 0.5486  0.6809[/B]  0.4514  0.5419  0.5129 [B] 0.2163  0.5523[/B]  0.7372  0.5339   
Games in this group: 34
0.7647  [B]0.8235  0.8214[/B]  0.1765  0.5     0.5588  [B]0.0588  1 [/B]      0.9412  0.5312  9.5294

NO Saints -3 vs NYG
This game is weird, talking to sports fans it seems the Giants will win this easily. My system does not have them winning outright in any of the groupings, and only one where the Saints fail to cover. I have included several different predictors as this game has several grouping systems where the games in the group higher than 10.

I really like the results of predictor two, where it shows very good results in three different groupings with different prediction tendencies. In the three groupings for the network, the percent of home wins predicted varies from .37 to 1, with accuracy ranging from .69 to .85. The fav predict percent varies from .25 to .66, with accuracy from .625 to .77.
Predicter 1: 14

Acc      HPred   H Acc   AwPred  Aw Acc  Sp Acc  FvPred  FavAcc  DogPred DogAcc  Pts Missed avg
0.6257  [B]0.6529  0.6578[/B]  0.3471  0.5654  0.5125 [B] 0.2336  0.5582[/B]  0.7168  0.5337  
Games in this group: 35
0.5714  [B]0.7714  0.5185[/B]  0.2286  0.75    0.5294  [B]0.2571  0.625[/B]   0.6571  0.5652  11.9143

Predicter 2: 13

0.628   [B]0.6166  0.6689[/B]  0.3834  0.5621  0.5113  [B]0.2733  0.5535[/B]  0.6699  0.5361 
Games in this group: 35
0.6857 [B] 0.3714  0.6923[/B]  0.6286  0.6818  0.5294 [B] 0.2571  0.625[/B]   0.6286  0.5909  11.0571 
Games in this group: 17
0.7647 [B] 0.9412  0.75[/B]    0.0588  1       0.625  [B] 0.5294  0.625 [/B]  0.4706  0.625   8.1765  
Games in this group: 21
0.8571  [B]1       0.8571[/B]  0       0       0.7     [B]0.6667  0.7857[/B]  0.2857  0.5     9.5238

Predicter 3: 14

0.6299  [B]0.735   0.643 [/B]  0.265   0.5934  0.5172 [B] 0.2877  0.5455[/B]  0.6688  0.5373  
Games in this group: 38
0.6579  [B]0.8947  0.6765[/B]  0.1053  0.5     0.6571  [B]0.3421  0.7692[/B]  0.6053  0.619   12.0789

Predicter 4: 14

0.6072 [B] 0.4771  0.6965[/B]  0.5229  0.5257  0.5133  [B]0.2284  0.5476[/B]  0.7297  0.5313  
Games in this group: 13
0.9231  [B]1       0.9231[/B]  0       0       0.5385  [B]0.4615  0.8333[/B]  0.4615  0.3333  13.5385

I have not placed any bets yet, but plan to watch the lines and do so over the next few days. Thanks, and good luck!

[Smogs]

I don't have a clue what the numbers mean but the majority of your picks look good, i especially like kansas, atlanta and seattle

[Indecent]

Thanks for you insight. I'll be editing the post to try to clarify what each of the numbers means...

[solobass]

nice start brother. what you will find (what i have found) is that to develop deadly accuracy (i am 69% on the year and increasing) is how do you weight the data. what is more important, what is least important and how to weight accordingly. best of luck!

[Indecent]

solo- Thanks for the advice and encouragement! It's a long and tough road, but the work should pay off in the long run!

Even though I would consider the system done for this year, I completely agree with the importance of stats used and I'm still experimenting with how each stat X is represented, (X per play, X per point, X per minute of possession, etc) and which stats (pass yards, interceptions, penalty yards, etc) are included in the calculation. Fortunately, once I have decided which stats to include, the weighting of these values is handled by the training program, so I only need to be aware of which values are not used (or hardly used) by the predictor.

When you mention 69%, is that ATS? Even if not, it's very impressive!

Best of luck in the future!

[Indecent]

I'm posting in here rather than creating a new topic that isn't relevant to most anyway.

Question to all computer cappers/modelers... What kind of average points missed are you hitting? Looking at previous results from a computer prediction tracker website it seems 10.5-12 is good, with spread being around 12-13 (open vs close, etc). Anyone hitting closer to 10 or into 9 long term? What kind of accuracy are you getting at predicting the straight up winner?

[rk9]

very cool.

One thing about the Giants vs. Saints game is that the Giants tend to be a very good road team. Personally I wouldnt say the Giants will win easily (like your friends said), I think it is a no play or a slight lean to the Giants. Of all the teams, the Giants are a team the numbers might underrate and the Saints are a team the numbers may tend to overrate.

[Indecent]

Thanks for your insight on how representative the stats are of each team! I've been so busy writing code for this and work that I haven't had much time to watch the games (or even highlights!).. I'll probably take your advice to heart and pass on the NO/NYG game, but I'm interested to see how it does!

Edit: I've had a couple people send me the same PM question, so I figured I could just post it here for everyone. The idea of this is simple...

Imagine a collection of friends who are all competent handicappers, but all look at different aspects of the game. Some might pay attention to the importance of home field on monday night, some might focus on yards per game, another on defensive stats per game, etc, etc. On their own, it isn't even important if they are long term winners, as long as you know which type of games they are long term winners.

So, we know which games they are good at, and when a game comes up that suits their skill set, we can use their opinion and ignore the others who don't specialize on this game type.

That's the idea at least...

[Indecent]

I've been reading others opinions and going through the results of my system and have some bets placed for this week.
KC +6.5
Seattle -3
Jets -9.5
Cinci -4.5 (moved to -5.5 now, woooo!)
Denver +3.5 (moved to +3, woooo!)
Min -3

There will be probably be more coming, waiting on few line changes and weather reports.

Best of luck everyone!

[Indecent]

I've been short on time this week so I couldn't get in my analysis, but here are the bets I made..

NO -6
GB -8.5
IND -14
Min +6
San Fran +3
ARI +7

GL all!