Lacing Them Up: Machine Learning and Basketball

The NBA season is back in action, and there’s already plenty of intrigue. The defending champion Golden State Warriors limped off to a 1-2 start, prompting far-too-early speculation that they’ve lost their magic touch. LeBron James and Dwyane Wade joined forces once again, this time in Cleveland. And the Phoenix Suns did their best impersonation of a video game by all running exactly at once after a turnover:

Such a well-oiled machine, huh? That fluid movement actually segues nicely into another topic the NBA has been working towards: using machine learning to enhance the product on the court.

As of the start of this season, Second Spectrum is now the official optical tracking provider of the league. The company has installed a player-tracking system in NBA arenas that uses a unique technology for machine understanding to provide a comprehensive analysis of players and teams.

Second Spectrum’s technology offers advanced statistics like speed, distance, drives, paint touches, and defensive impact. Let’s take a look at some of the applications this machine learning provides.

Shot Charts

A team can quickly analyze a shot chart like the one below to see where they’re most efficient. Defenses could also study these patterns to see where players spend the most time on the court. If a player shoots a better percentage from the right corner than from the right wing, he’ll naturally want to hang out more in the right corner. But an alert defender armed with that knowledge can stay one step ahead by forcing his man to cut towards the wing or the center of the court, rather than staying in the corner.

Second Spectrum Heat Map

That could be the difference between winning and losing. A corner three-pointer provides 30 more points per 100 possessions than a three-pointer from the wing does. As most teams are averaging more than 100 possessions this season, that’s a big deal!

Defensive Tendencies

Player tracking can also monitor what players do defensively and how it affects their team. Does a certain player defend the pick and roll by fighting over the top of a screen while his teammate steps out to help before returning to his man? An offensive player could adjust by popping out for an open three-pointer since the defender will be a step late in coming to guard him.

Player Touches

Does it ever seem like certain players hold the ball for a long time, dribbling a lot before passing or shooting? In some cases, it’s because they are doing just that. Last season, Washington Wizards Guard John Wall averaged more than six seconds per touch, dribbling nearly six times every time he touched the ball. Defenses can expect him to dribble and drive when he gets the ball and adjust their game plan accordingly.

Conversely, the Oklahoma City Thunder’s Russell Westbrook and the Toronto Raptors’ DeMar DeRozan were the only two players to shoot more than 10 shots per game on dribble pull-ups. Those types of shots accounted for more than half of their field goal attempts every game. A wise defender would play an extra step up when those players come down the court to try and cut down on some of those attempts.

The NBA has been at the forefront of modern technology for years. It began using player-tracking technology back in 2009 and became the first professional sports league in the United States to use the technology for every single game in 2013. We’ve seen a glimpse of how basketball and machine learning can co-exist, but as these processes continue to improve, the sport will be even more analyzed and broken down. For stat geeks, it’s an absolutely thrilling time.

The use of technology in this endeavor ranges widely to meet the requirements for infrastructure, creative, application, and data science needs. Applications like Second Spectrum’s have a need for highly scalable, fault-tolerant, low-latency systems that can bring together many data feeds and visualize data and process and serve high-quality video content.

Second Spectrum’s tech stack not only includes AWS and Route 53 but also Google Universal Analytics and application containerization like Docker along with machine learning applications. All of these combine for a platform of creative big data and machine learning needs.

To read more on how companies can leverage machine learning check out my colleague Patrick McClory’s recent blog post, The Real Value of Machine Learning. Even off the court, machine learning can help bring any enterprise to the next level. And that’s just as exciting as any buzzer beater.

Datapipe has joined forces with Rackspace to create the world’s leader in multi-cloud managed services. Learn more about the acquisition here.

David Lucky is a Product Marketing leader at Rackspace for the Managed Public Cloud services group, a global business unit focused on delivering end-to-end digital transformation services on AWS, Azure, GCP and Alibaba. David came to Rackspace from Datapipe where as Director of Product Management for six years he led product development in building services to help enterprise clients leverage managed IT services to solve complex business challenges. David has unique insight into the latest product developments for private, public and hybrid cloud platforms and a keen understanding of industry trends and their impact on business development. He holds an engineering degree from Lehigh University and is based out of Jersey City, NJ. You can follow David on LinkedIn at and Twitter @Luckys_Blog.