What can a machine tell us about sports?

by thesanction1

Machines don’t feel.  They don’t comprehend, they don’t react, they don’t compete and they don’t understand.  They do only what we allow them to do.  So how is it that we allow them to do so much more than we might be able to do ourselves? 

When it comes to sports, its easy to resist the urge to “crunch numbers” in order to predict or understand events.  We think of sports as such a human affair – really knowing a sport requires understanding the rules, the history, the key concepts and goals associated with each game. In the course of any sporting event there are strategies playing out, injuries taking place, character being tested, and large amounts of creative juices flowing as us humans attempt to triumph over one another, or even over ourselves.  

I get it, trust me.  We aren’t numbers, sports aren’t algorithms. There is an element of sport that will forever be impenetrably and inextricably bound by a vale of the truly magical, the unexpected and the seemingly impossible.  This allure is what draws us in and makes us so fanatically obsessed with sports. The obsession can grow to the point that it we feel compelled to place that which we hold most sacred (cold hard cash, obviously) on the line in the hopes that we might be better than the next guy at understanding all these wonderful dynamics, and be able to profit by predicting how they play out.  You see, my friends, there’s money to be made in predicting the outcome of ‘sport’. And where there’s money to be made, some nerd like myself will try to find a way to use computers to make it.

Thus the purpose of this blog and the topic of this first post.  I want to share with all those who will take notice the predictions I’m making about sports (particularly football, but when football season is over I’ll turn my attention towards golf).  I’m using the most advanced methods available, with custom algorithms and curated data sets, looking at unique variables I’ve concocted over these last few years, to produce what I consider to be the most accurate predictions for team and individual performance in the NFL ever to be put into a blog and given freely to the public.   

In short – and as idiotic as this sounds – I think I can beat the odds makers.  Correction: I have evidence that I can beat the odds makers.  I know I can help your fantasy football teams.  As the saying goes – the proof is in the pudding.  So I’m laying out a big hot steamy pile of fresh pudding for everyone to see, touch, and taste week after week, till their stomachs hurt from all the rich puddingly-goodness.  Once a post is made, it will not be altered.  Each week I will review my picks, and try to explain where the algorithms went wrong, and what it got right.  At the end of the season I hope to have a chain of blog posts which testifies to the effectiveness (or lack thereof) of advanced algorithmic approaches to predicting sports outcomes.  Without further ado – let the games begin. 

The following are all projected outcomes, against the spread, for the week 11 NFL match-ups:


These picks were made as of Saturday, November 16th 2013 using data that was updated on Thursday afternoon before the evening games.  The Thursday game prediction is valid, though I will not be including it with any more than a mention in my posts. Obviously predicting something correctly after it’s already happened isn’t the point. For my own record keeping, the above predictions were made using machine learning techniques, rather than traditional linear methods or iterative linear model search techniques. 

The next projections I’ll include are for fantasy purposes.  The following are the projected fantasy point production totals for QB’s. The projections here reflect http://www.DraftStreet.com’s scoring standards (four points for passing TD’s, 6 for rushing, and half point PPR which obviously doesn’t come into play for QB’s).  These projections were made using an iterative linear modeling method:


I have made predictions for the over-under on each matchup, as well as fantasy point predictions for RB’s, WR’s, TE’s and K’s which I use in my own personal fantasy team creation and sports betting endeavors … but for the blog I will limit the presentation to the two heaviest hitters: match-ups against the spread and QB’s.  

Briefly some comments on both sets of projections…

Comments on spread projections

The key column in understanding the spread projections is the “SpreadCoverIndex” column.  The number here represents two things.  First, if the sign is positive it implies that the away team will cover the spread. By deduction the cleverer among you have no doubt inferred that a negative sign implies that the home team will cover the spread.  The size of the number represents how much the model predicts the home or away team will cover by, on average, if the game were played over an over again ad infinitum.  So a bigger number ought to imply a more confident prediction. As an example – the algorithm picked the Titans correctly to cover +3  against the Colts in the Thursday night match-up, which would have resulted in a 27-30 push. 

Comments on QB fantasy point projections

I think the idea here is pretty obvious. If Drew Brees were to play this weekend over and over again, the average amount of points he would score is projected to be 24.25.  Obviously you can ignore predictions of negative points – this doesn’t so much imply a that a Brock Osweiler will be worth negative points this week, or even on average. It is more a remnant of the linear model being applied to a player whose stat line is so terrible or underpopulated that the model projects him to score negative points.  In essence all negative projections produced by a linear modeling method can be considered projections of 0 total points, unless, of course, those projections are for Eli Manning.  

Ok, that’s all for this week. Good luck to everyone, whatever they are doing on Sunday.  I, for one, will be celebrating the holiest of all holy days, gorging myself on Red Zone and IPA’s, enjoying laughs and heartburn with friends and, as always, pulling for a world in which I can teach machines to teach me a little something about sports.