It amazes me that football hasn't embraced sabermetrics like baseball has. Not going for it on a 4th and anything under 5 is basically stupid. Possess the ball, score as many touchdowns as humanly possible, and be aggressive.
As a data scientist, I fully agree. When every aspect of the game can now be categorized & analyzed, data based decisions, schemes, play calls, formations, etc will only increase the probability of success.
Its why I hope Brown hires an actual data scientist as an analyst. While its good to have a few coaches/ex-players as "analyst", they can only do so much, they cant extract the full value from data.
If its <= 4th & 6, and you are more then 30-35 yards from your own end-zone, every coach should go for it.. Thats a very small part of it tho... When you can break down every aspect of historical games, offensive schemes vs Defensive schemes, formations, routes, motions, RPO's, etc, applying machine learning is the wave of the future.. Machine learning & predictive models can provide a coach or OC/DC with more insight into how to be less predictable and increase the odds of positive outcome. An ex-coach/player can only extract probably 25-30% of the value from advanced data analytics...
I would love to see WVU leverage its data science engineering dept and develop a genetic optimization & machine learning model that can be used for two purposes. 1 Pre-game scheming & 2. In-game real-time "Next Play" probability.
pre-game scheming model would output probabilities for how the team evolves week to week, likely new plays implemented, best concepts, formations, schemes, matchups, etc, best reactive schemes, play setups, option routes, blocking concepts etc... The model would be updated with the last game's data, and be ready on Sunday for the coach. A detailed decision tree of information the coach can use when developing his weekly game plan.
"Next Play probability" can be split into two prediction categories. 1 the most likely call/play from opponent 2. Optimal play call for the top 4-5 most likely opponent "reaction/decisions"...
Smart tablets can be connected to a VPS that runs the real-time models, and real-time in-game data can be feed into the model for the most accurate "predictions".. Before a coach calls the next play, he looks down and sees a probability map of probable opponent calls & optimal plays to counter(or plays/formations/etc that will most likely succeed). A coach can then use his "feel" for the game to select which of those he feels is best, given player matchups, opponent weaknesses, etc..
This wil happen at some point in the future. Its only a matter of time. The days when coaches called plays on their own will be looked at as archaic.. Brown does use data analytics more then DH, which I like, but I really hope he hires someone who can provide additional insights, a unique perspective & tools that only a data scientist can provide. It would be far more valuable and worth every penny... The variables included in these models can be so specific and down to opponent matchups, player speed, height, weight, down & distance, time remaining in each qtr, home/away, time of day, weather conditions, side of field, wind, in addition to all the data from every game the opponent has played, every game a coach has called, all formations, all blocking schemes, all routes, QB patterns, etc... The more variables, the more accurate the predictions can be... While a ex-player analyst can provide generalities or likelihoods, a human cannot match the specificity of a probability engine, and when you combine that with the "fuzzy logic" humans excel in, and a coaches feel for the game, it would be a great combo.. I would much rather WVU be on the front end of this and have it as an edge for a few years and not lag behind programs who come around sooner... When it happens, I think you'll see a major shift in the game. You'll see many new alternative formations, OC&DC concepts, and programs fighting over the best data scientist... 10+ years from now, they will be recruited like 4-5 stars and HC's are now...