At this point, I’m sure you’ve heard that data is taking over the world. It’s already old news that we have been generating more data than ever before, and that this number is growing exponentially. It has already reached most areas, and it’s not stopping anytime soon. Every area of our lives can be improved with the study of data.
Right now, let’s focus on sports. More specifically: Basketball. Since the beginning, stats have been an important part of sports. Boxscores, scouts, and others have been around for a while. But that’s just the first step on a thousand-mile journey. Here, we’re going to analyze how Machine Learning and Data Science can and have impacted the game of Basketball.
First of all, I must point out that this is not a coding tutorial or a guideline for Data Science projects. Rather, it’s a conversation, exposing the numerous possibilities Machine Learning brings to sports. Think of it as food for thought.
With that out of the way, let’s get started.
Let’s start with the basics. On a Data Science Project, exploratory data analysis plays a huge role. It shouldn’t be different when approaching sports. Understanding each and every aspect of the game, while being concerned with how it translates to reality, must be the first step. In the NBA, there’s something called Advanced Stats. Regular old boxscores just don’t cut it anymore. The result is the creation of new metrics upon which we analyze how these affect the outcome of games. And these are being created and analyzed regularly.
After thoroughly analyzing your data, there are some other things you can try out. Kaggle has a yearly Machine Learning competition sponsored by the NCAA and Google Cloud on Men’s Basketball NCAA Tournament. The goal is to predict the outcome of the tournament game by game, revealing potential upsets, underdogs, Cinderella teams, and whatnot.
They provide Data Scientists with historic data of the tournament, on several different variables. Basketball teams can use real data from the championships they’re in, and use a predictive model to test how well they would rank in a certain championship if they could improve such and such attributes, or add this or that player to the roster. It’s great for scouting, game prep, evaluating the effort of a given tactical strategy and even better choosing what players should try to develop or improve in their games. Again, food for thought.
All of the previous choices sound great and work wonderfully. However, if you really want to reach the next level, this is the goal. Computer Vision is taking over the world, and it’s already a thing in sports. Kinexon holds the majority of the market in the NBA and provides state-of-the-art solutions for teams. Real-time player assessment, fatigue prediction, measurement of new variables on demand. This is your goal if you’re serious about taking Data Science and Machine Learning to sports. With this kind of technology applied to your team, working for you and your personal needs, this will feed the previous steps of your pipeline and help you and your team reach the next level.
If you’re a basketball fan, you’ve probably seen this picture on the internet these days. It shows the evolution of shot selection in the NBA. We have to give credit to the Houston Rockets on pioneering the heavy use of Data Science and Analytics in the NBA.
In simple words, Analytics has shown that the 2-point mid-range jumper is a bad shot and that you’re better off shooting threes or taking it to the rack and getting an easy layup or a foul. Watch a Houston Rockets game and you’ll see it to perfection (Thank you, James Harden). Players are shooting more and more threes from deeper and deeper. The game is completely different, and we have analytics to blame (or thank?). It’s all fun and games until it hits you. Carmelo Anthony has played for the Houston Rockets and it almost cost him his career. Being dominant in the so-called Iso Era of the NBA, Melo heavily relied on his mid-range jumper to score. After 10 days with the team, they waved him. A 10-time all-star, elite-level player almost retired because of what analytics showed about him. If you’re interested in this story, I recommend this post.
Fortunately, he bounced back, found a spot in the Portland Trailblazers and recently hit a game-winning shot, which, ironically, was a mid-range 2-point jumper. At the 2:00 mark, you can see the shot for yourself.
Absolutely not. But it’s not everything. Thoroughly analyze your data, scout your players, use all the models in the world, but always watch the games, see for yourself if your data is telling the whole story. With Computer Vision we can measure stats that could never be measured, but we’ll always have that player that has “something the stats don’t show”.
This was just a brief talk on the subject, and as I said, food for thought. I hope you found this interesting and insightful, and should you have any questions or just want to talk basketball, make sure to connect with me on LinkedIn and check out my projects on GitHub.
For more reading on Machine Learning and Basketball, I recommend this post by Towards Data Science.