Conquering Fantasy with Data Science
Football is back this week (!!!!!) and with it comes the obsession of fantasy football. I myself am taking part for the very first time — I’m not a gambler; I’m cosmically unlucky — but as I’m staring at my team for this upcoming week I’m quickly going down a rabbit hole of how to make the most with all the information I have at my fingers tips to make the best roster I can and crush my opponent — despite being cosmically unlucky, I’m beyond competitive and now must win.
Fantasy football has taken on a life of its own. From marathon drafts before the season starts to the weekly maintenance to the at time humiliating consequences of coming in last, fantasy is a great way to connect to others as well as professional football. You feel a part of the game when you have a stake in all the games and not just your home team. Fantasy has gone beyond just understanding who is injured and out for the week to a serious investment of time and resources.
Succeeding at fantasy means you basically can predict the future or you’re crazy lucky. But some people are finding a way to crack the fantasy code using data science. The data is out there so with the right tools you can step in and become a digital football genius. If only it came with Patrick Mahomes’ paycheck.
So how exactly are people using data science to win at fantasy? Where historically one was using a player’s average as an indicator of future performance, one could now layer in the nuances that come up in the ever changing landscape and pinpoint the most relevant features to attempting to predict the efficiency of a player.
One of the big focuses on in making use of data is around opportunities or how much a player is being targeted. A wide receiver or tight end is a quick way to amass points in the game. The more a player is targeted, the more likely they’ll return points. Quarterbacks understandably develop relationships with their receivers and target some more than others. These bonds usually result in more point conversions — think Tom Brady and the newly un-retired Gronkowski. Having a highly targeted receiver on your fantasy team boosts your points earning ability. There’s also this notion that simply looking at the number of catches doesn’t fully capture the story on the field. A player can be targeted a number of times and not bring the ball in for whatever reason. That’s not to say they’re have a poor return rate on those targets, but that there is more context to understand what was happening on the field.
Getting data on a football game means really getting into the weeds of game stats. There’s a lot of moving parts with previous games while also bearing in mind what’s happened the week leading into the game. It can be difficult to isolate what features will be most important in an algorithm and set up what circumstances will change that feature importance. First & 10 looks a lot different than 3rd & 1 and changes the level players compete at in a given moment.
One big issue with trying to predict football results is the tightness of the season. Where baseball has over 160 games in its regular season and even basketball and hockey who play 82 regular season games, football only has 16 games. It’s really hard to develop and trust trends in such a small data pool.
Overall there’s a lot of potential when it comes to leveraging data science to win at fantasy football. There’s a lot of untapped potential with algorithms and different perspectives will continue to evolve the industry. I look forward to seeing what I can do with my own data science skills in my league. They should fear me and my computer.
Source:
