Created machine learning program to calculate a score for each football team from 1950 to the present. The program first uses ELO to score each team for each game since 1869. The ELO rules used were copied from a standard chess ELO (start @ 1500, learning 'k' factor or 32...). The program calculates a per game ELO average for each team over the last 25 seasons (ex 1950 season computation uses every game from 1925-1950 to score an average ELO for each team) with a slight bonus for the most recent 5 seasons. ELO is the major component of the score, but it also includes a bonus for teams making and winning bowl games as well as bonus for being ranked in the AP poll and a small bonus for strength of schedule (1% of opponent ELO regardless of who wins the game).

The goal was to determine which teams could be considered 'blue bloods' each year and to find when teams elevated and fell from 'blue blood' status. Not sure how to quantify 'blue blood' in the scoring so made the top 10 ranked teams each season a 'blue blood'.
Ex.) Notre Dame was a 'blue blood' every year from 1950 until 2002 where it began falling from the top ten teams until #20 this season.
Ex.) Florida State is one of the worst teams in the 1950s and steadily climbs to 'blue blood' status in the 1990s.

The following spreadsheet captures the results for each year.
https://docs.google.com/spreadsheets/d/1...sp=sharing

I hope it isn't a finished product. You have the 1959 national champion listed 65th, one behind a team they beat 71-0.

Note that this is not a single year ranking, it accumulates the results from the previous 25 seasons. While winning a championship one year will score very high in one season if the previous 24 seasons were average then the score will remain average. Margin of victory is not included in the scoring.

Notre Dame had quite a run as #1, even when they had losing seasons.


Sent from my iPad using Tapatalk

(01-03-2019 12:59 PM)TexanMark Wrote:  I hope it isn't a finished product. You have the 1959 national champion listed 65th, one behind a team they beat 71-0.

And Notre Dame finished 5-5 that year, but still is #1 on his sheet.

So much, much work is needed.


Not sure how a team did in past years (especially more than a year) affects how a team does in the current year.

(01-03-2019 01:08 PM)dbackjon Wrote:  

(01-03-2019 12:59 PM)TexanMark Wrote:  I hope it isn't a finished product. You have the 1959 national champion listed 65th, one behind a team they beat 71-0.

And Notre Dame finished 5-5 that year, but still is #1 on his sheet.

So much, much work is needed.


Not sure how a team did in past years (especially more than a year) affects how a team does in the current year.

please read original post, this rank is NOT a single season ranking method, it uses MANY previous seasons to score each team. THe goal was to determine which teams were consistently good over time. Looks at when Notre Dame score began to decline correlates to when then began to be inconsistent for many seasons.

(01-03-2019 01:14 PM)jhasting Wrote:  

(01-03-2019 01:08 PM)dbackjon Wrote:  

(01-03-2019 12:59 PM)TexanMark Wrote:  I hope it isn't a finished product. You have the 1959 national champion listed 65th, one behind a team they beat 71-0.

And Notre Dame finished 5-5 that year, but still is #1 on his sheet.

So much, much work is needed.


Not sure how a team did in past years (especially more than a year) affects how a team does in the current year.

please read original post, this rank is NOT a single season ranking method, it uses MANY previous seasons to score each team. THe goal was to determine which teams were consistently good over time. Looks at when Notre Dame score began to decline correlates to when then began to be inconsistent for many seasons.

The main issue I see is that by including that degree of past data it is difficult to spot a significant rise or decline. For example, if Florida State continues to bottom out and I'm looking at this a decade from now I'm thinking that they were pretty good in 2016 and 2017, when in fact those years would be the start of their downward trend.

(01-03-2019 01:28 PM)RobUCF Wrote:  

(01-03-2019 01:14 PM)jhasting Wrote:  

(01-03-2019 01:08 PM)dbackjon Wrote:  

(01-03-2019 12:59 PM)TexanMark Wrote:  I hope it isn't a finished product. You have the 1959 national champion listed 65th, one behind a team they beat 71-0.

And Notre Dame finished 5-5 that year, but still is #1 on his sheet.

So much, much work is needed.


Not sure how a team did in past years (especially more than a year) affects how a team does in the current year.

please read original post, this rank is NOT a single season ranking method, it uses MANY previous seasons to score each team. THe goal was to determine which teams were consistently good over time. Looks at when Notre Dame score began to decline correlates to when then began to be inconsistent for many seasons.

The main issue I see is that by including that degree of past data it is difficult to spot a significant rise or decline. For example, if Florida State continues to bottom out and I'm looking at this a decade from now I'm thinking that they were pretty good in 2016 and 2017, when in fact those years would be the start of their downward trend.

Notre Dame is an example to your question, they were clearly a 'blue blood' in the 70s, 80s and 90s and then began to decline in the 2000s, you see see 10 years later that their score continues to fall every season. This is an alternate scoring method to the standard scoring, it does not and should not indicate if a team is good in a single season, it shows a smoothed average over time. This seemed like a good approach to determine a 'blue blood' since these are teams that are consistently good over many seasons.

(01-03-2019 12:57 PM)jhasting Wrote:  Created machine learning program to calculate a score for each football team from 1950 to the present. The program first uses ELO to score each team for each game since 1869. The ELO rules used were copied from a standard chess ELO (start @ 1500, learning 'k' factor or 32...). The program calculates a per game ELO average for each team over the last 25 seasons (ex 1950 season computation uses every game from 1925-1950 to score an average ELO for each team) with a slight bonus for the most recent 5 seasons. ELO is the major component of the score, but it also includes a bonus for teams making and winning bowl games as well as bonus for being ranked in the AP poll and a small bonus for strength of schedule (1% of opponent ELO regardless of who wins the game).

The goal was to determine which teams could be considered 'blue bloods' each year and to find when teams elevated and fell from 'blue blood' status. Not sure how to quantify 'blue blood' in the scoring so made the top 10 ranked teams each season a 'blue blood'.
Ex.) Notre Dame was a 'blue blood' every year from 1950 until 2002 where it began falling from the top ten teams until #20 this season.
Ex.) Florida State is one of the worst teams in the 1950s and steadily climbs to 'blue blood' status in the 1990s.

The following spreadsheet captures the results for each year.
https://docs.google.com/spreadsheets/d/1...sp=sharing

Interesting work.

I do find the Texas and Nebraska results in the 2000s odd. Nebraska is rising during a slump for them. Texas dropped from 1998-2009 from 11 to 15 despite having the 2nd best record in that era behind Ohio St.

From 1978-1992, 9 schools, Ohio St., Michigan, Notre Dame, Penn St., Texas, Oklahoma, USC, Alabama and Nebraska, controlled the top 8 slots. If you include Tennessee and the Florida schools, who all moved into the top 12 in 1995, it was 2015 before anyone other than those 13 occupied one of the top 8 slots (Georgia moved in).

A real good example is Boise State, they start way down in the 1990s and make a slow climb to the lower 20s presently.


Sent from my iPad using Tapatalk

Auburn and Ohio State split the national championship in '57. Ohio State is 6th and Auburn isn't even on the list probably because the computer didn't recognize Alabama Polytechnic Institute, or API.

Alabama is ranked in the top 2 during Ears Whitworth's worst ever Alabama teams, including their 3 win season.

(01-03-2019 01:59 PM)Jjoey52 Wrote:  A real good example is Boise State, they start way down in the 1990s and make a slow climb to the lower 20s presently.


Sent from my iPad using Tapatalk

In the other direction, you see Army and Minnesota make a slow decline from being #6 and #9 in 1950. Similarly, Georgia Tech makes a gradual (but not immediate) decline after leaving the SEC in 1964. Penn St. was #49 in 1950 but moved up to #5 by 1978. OU was #21 in 1950 but had their run in the 50s and moved up by 1959 to #3. Texas moved from #12 in 1950 up to #1 by 1967 in the Darrell Royal era. Nebraska had a long slump in the 40s and 50s and dropped from #25 in 1950 to #51 before rapidly moving up in the 70s&80s to #14 by 1978 and #9 by 1983.

(01-03-2019 02:04 PM)JRsec Wrote:  Auburn and Ohio State split the national championship in '57. Ohio State is 6th and Auburn isn't even on the list probably because the computer didn't recognize Alabama Polytechnic Institute, or API.

Alabama is ranked in the top 2 during Ears Whitworth's worst ever Alabama teams, including their 3 win season.

Auburn is #40 in 1957. The program does show a dominant 1957 season for Auburn and they improved almost 100pts in ELO, but again this method is NOT ranking single seasons, looking at Auburn from 1932-1957 shows many good seasons (they are #20 in 1938) but also many bad seasons until they begin consistently winning in the 70s. This is evident because they go from #61 in 1950, to the 30s through the 1960, then the teens in the 70s where they remained until this season.


ELO,Auburn,1978.90464209402,WON,Auburn,7,@,<8> Tennessee,0,Sep 28 1957,
ELO,Auburn,1981.56926717746,WON,<7> Auburn,40,vs,Chattanooga,7,Oct 5 1957,
ELO,Auburn,1997.64373173413,WON,<9> Auburn,6,vs,Kentucky,0,Oct 12 1957,
ELO,Auburn,2020.71670062014,WON,<9> Auburn,3,@,Georgia Tech,0,Oct 19 1957,
ELO,Auburn,2029.31965089445,WON,<5> Auburn,48,@,Houston,7,Oct 26 1957,
ELO,Auburn,2040.87783551551,WON,<4> Auburn,13,vs,<19> Florida,0,Nov 2 1957,
ELO,Auburn,2049.75700820969,WON,<3> Auburn,15,vs,<17> Mississippi State,7,Nov 9 1957,
ELO,Auburn,2055.42171200981,WON,<3> Auburn,6,vs,Georgia,0,Nov 16 1957,
ELO,Auburn,2057.63910744007,WON,<2> Auburn,29,@,Florida State,7,Nov 23 1957,
ELO,Auburn,2062.20257652763,WON,<1> Auburn,40,vs,Alabama,0,Nov 30 1957,

Here is an example where the look back is only 10 seasons. As expected this is much less stable and shows faster transitions. Ex.) Look a Nebraska at #1 in 1992 and stay 1,2, or 3rd until 2008 when they begin to fall quickly to #22 this season.

https://docs.google.com/spreadsheets/d/1...sp=sharing

And finally an example with no look back. This correlates to the end of year AP ranking and does not demonstrate winning consistency. Ex. look at Alabama 2005-2008 rank drop into the 30s range because of a few bad seasons and then their score recovers quickly to #1 post Shula.

https://docs.google.com/spreadsheets/d/1...sp=sharing

(01-03-2019 02:17 PM)jhasting Wrote:  

(01-03-2019 02:04 PM)JRsec Wrote:  Auburn and Ohio State split the national championship in '57. Ohio State is 6th and Auburn isn't even on the list probably because the computer didn't recognize Alabama Polytechnic Institute, or API.

Alabama is ranked in the top 2 during Ears Whitworth's worst ever Alabama teams, including their 3 win season.

Auburn is #40 in 1957. The program does show a dominant 1957 season for Auburn and they improved almost 100pts in ELO, but again this method is NOT ranking single seasons, looking at Auburn from 1932-1957 shows many good seasons (they are #20 in 1938) but also many bad seasons until they begin consistently winning in the 70s. This is evident because they go from #61 in 1950, to the 30s through the 1960, then the teens in the 70s where they remained until this season.


ELO,Auburn,1978.90464209402,WON,Auburn,7,@,<8> Tennessee,0,Sep 28 1957,
ELO,Auburn,1981.56926717746,WON,<7> Auburn,40,vs,Chattanooga,7,Oct 5 1957,
ELO,Auburn,1997.64373173413,WON,<9> Auburn,6,vs,Kentucky,0,Oct 12 1957,
ELO,Auburn,2020.71670062014,WON,<9> Auburn,3,@,Georgia Tech,0,Oct 19 1957,
ELO,Auburn,2029.31965089445,WON,<5> Auburn,48,@,Houston,7,Oct 26 1957,
ELO,Auburn,2040.87783551551,WON,<4> Auburn,13,vs,<19> Florida,0,Nov 2 1957,
ELO,Auburn,2049.75700820969,WON,<3> Auburn,15,vs,<17> Mississippi State,7,Nov 9 1957,
ELO,Auburn,2055.42171200981,WON,<3> Auburn,6,vs,Georgia,0,Nov 16 1957,
ELO,Auburn,2057.63910744007,WON,<2> Auburn,29,@,Florida State,7,Nov 23 1957,
ELO,Auburn,2062.20257652763,WON,<1> Auburn,40,vs,Alabama,0,Nov 30 1957,

This system may be fine for tracking a long range trend, but for rating schools by season it is beyond absurd and worthless.

(01-03-2019 03:14 PM)JRsec Wrote:  

(01-03-2019 02:17 PM)jhasting Wrote:  

(01-03-2019 02:04 PM)JRsec Wrote:  Auburn and Ohio State split the national championship in '57. Ohio State is 6th and Auburn isn't even on the list probably because the computer didn't recognize Alabama Polytechnic Institute, or API.

Alabama is ranked in the top 2 during Ears Whitworth's worst ever Alabama teams, including their 3 win season.

Auburn is #40 in 1957. The program does show a dominant 1957 season for Auburn and they improved almost 100pts in ELO, but again this method is NOT ranking single seasons, looking at Auburn from 1932-1957 shows many good seasons (they are #20 in 1938) but also many bad seasons until they begin consistently winning in the 70s. This is evident because they go from #61 in 1950, to the 30s through the 1960, then the teens in the 70s where they remained until this season.


ELO,Auburn,1978.90464209402,WON,Auburn,7,@,<8> Tennessee,0,Sep 28 1957,
ELO,Auburn,1981.56926717746,WON,<7> Auburn,40,vs,Chattanooga,7,Oct 5 1957,
ELO,Auburn,1997.64373173413,WON,<9> Auburn,6,vs,Kentucky,0,Oct 12 1957,
ELO,Auburn,2020.71670062014,WON,<9> Auburn,3,@,Georgia Tech,0,Oct 19 1957,
ELO,Auburn,2029.31965089445,WON,<5> Auburn,48,@,Houston,7,Oct 26 1957,
ELO,Auburn,2040.87783551551,WON,<4> Auburn,13,vs,<19> Florida,0,Nov 2 1957,
ELO,Auburn,2049.75700820969,WON,<3> Auburn,15,vs,<17> Mississippi State,7,Nov 9 1957,
ELO,Auburn,2055.42171200981,WON,<3> Auburn,6,vs,Georgia,0,Nov 16 1957,
ELO,Auburn,2057.63910744007,WON,<2> Auburn,29,@,Florida State,7,Nov 23 1957,
ELO,Auburn,2062.20257652763,WON,<1> Auburn,40,vs,Alabama,0,Nov 30 1957,

This system may be fine for tracking a long range trend, but for rating schools by season it is beyond absurd and worthless.

Read the OP JR!!! That is what he is doing. He is tracking long range trends and not saying anything about season by season.

I do think there is some error in the Texas rankings in that 1998-2009 era. Texas should be rising, not dropping.

(01-03-2019 03:17 PM)bullet Wrote:  I do think there is some error in the Texas rankings in that 1998-2009 era. Texas should be rising, not dropping.

That does look peculiar, Texas is improving each year but their rank goes down or stays the same. Thanks for pointing out this example. This looks like a case where other schools supplant Texas. Here is an example where Texas improves their score from 2003 to 2004 but drops 3 spots because UGA USC and Auburn improve by a larger margin.

BB,Nebraska,2003,1,3708
BB,Florida State,2003,2,3634
BB,Miami FL ,2003,3,3531
BB,Florida,2003,4,3506
BB,Alabama,2003,5,3500
BB,Michigan,2003,6,3498
BB,Penn State,2003,7,3493
BB,Oklahoma,2003,8,3485
BB,Ohio State,2003,9,3440
BB,Tennessee,2003,10,3417
BB,Notre Dame,2003,11,3388
BB,Texas,2003,12,3364
BB,Georgia,2003,13,3363
BB,Southern California,2003,14,3348
BB,Washington,2003,15,3345
BB,Auburn,2003,16,3345

BB,Nebraska,2004,1,3707
BB,Florida State,2004,2,3687
BB,Miami FL ,2004,3,3566
BB,Oklahoma,2004,4,3533
BB,Florida,2004,5,3531
BB,Michigan,2004,6,3526
BB,Alabama,2004,7,3474
BB,Ohio State,2004,8,3471
BB,Tennessee,2004,9,3465
BB,Penn State,2004,10,3451
BB,Georgia,2004,11,3438
BB,Auburn,2004,12,3410
BB,Notre Dame,2004,13,3377
BB,Southern California,2004,14,3369
BB,Texas,2004,15,3368

(01-03-2019 03:16 PM)bullet Wrote:  

(01-03-2019 03:14 PM)JRsec Wrote:  

(01-03-2019 02:17 PM)jhasting Wrote:  

(01-03-2019 02:04 PM)JRsec Wrote:  Auburn and Ohio State split the national championship in '57. Ohio State is 6th and Auburn isn't even on the list probably because the computer didn't recognize Alabama Polytechnic Institute, or API.

Alabama is ranked in the top 2 during Ears Whitworth's worst ever Alabama teams, including their 3 win season.

Auburn is #40 in 1957. The program does show a dominant 1957 season for Auburn and they improved almost 100pts in ELO, but again this method is NOT ranking single seasons, looking at Auburn from 1932-1957 shows many good seasons (they are #20 in 1938) but also many bad seasons until they begin consistently winning in the 70s. This is evident because they go from #61 in 1950, to the 30s through the 1960, then the teens in the 70s where they remained until this season.


ELO,Auburn,1978.90464209402,WON,Auburn,7,@,<8> Tennessee,0,Sep 28 1957,
ELO,Auburn,1981.56926717746,WON,<7> Auburn,40,vs,Chattanooga,7,Oct 5 1957,
ELO,Auburn,1997.64373173413,WON,<9> Auburn,6,vs,Kentucky,0,Oct 12 1957,
ELO,Auburn,2020.71670062014,WON,<9> Auburn,3,@,Georgia Tech,0,Oct 19 1957,
ELO,Auburn,2029.31965089445,WON,<5> Auburn,48,@,Houston,7,Oct 26 1957,
ELO,Auburn,2040.87783551551,WON,<4> Auburn,13,vs,<19> Florida,0,Nov 2 1957,
ELO,Auburn,2049.75700820969,WON,<3> Auburn,15,vs,<17> Mississippi State,7,Nov 9 1957,
ELO,Auburn,2055.42171200981,WON,<3> Auburn,6,vs,Georgia,0,Nov 16 1957,
ELO,Auburn,2057.63910744007,WON,<2> Auburn,29,@,Florida State,7,Nov 23 1957,
ELO,Auburn,2062.20257652763,WON,<1> Auburn,40,vs,Alabama,0,Nov 30 1957,

This system may be fine for tracking a long range trend, but for rating schools by season it is beyond absurd and worthless.

Read the OP JR!!! That is what he is doing. He is tracking long range trends and not saying anything about season by season.

I play chess, play it well, and have never heard of an ELO outside of Jeff Lynne and the Electric Light Orchestra. Any system that is not self explanatory fails. And quite frankly old fading blue bloods will not want to be tracked and rated. New and rising programs will take a couple of decades to be recognized so they will skip it for quick publicity.

I grasped the concept in just looking at the data. It's probably good for a grad project in statistics but as far as having a broad application? Not so much.