Part 2: Getting the most out of NBA Lineups - Miami Heat

Embed from Getty Images

Previously I took my machine learning algorithm and found line-up specific insights for the Dallas Mavericks.

Post can be found here: http://www.zigzaganalytics.com/home/getting-the-most-out-of-nba-lineups-with-machine-learning

An example of an insight learned being how the Mavericks really struggle with Dirk at the 5 spot. Most notably the struggles come in terms of defensive rebounding and rim protection, but can cover for this by playing Dirk next to Dwight Powell as much as possible and be a positive +/-

For this post, I have taken a similar deep dive into Miami Heat line-ups utilizing machine learning and here are the results...

Key Thoughts:

The really interesting and surprising insight being the negative impact Goran Dragic is having on heat line-ups this season, with line-ups 2 & 4 both having a negative +/- to demonstrate this.

Dragic is a -2.3 +/- per game this season vs +0.9 last season and +2.8 the season before.

The downward trend is concerning.

The algorithm found to be of most concern, Dragic's inability to get to the free throw line and Dragic line-ups not getting up 3's at a rate that Miami's offensive strategy requires.

Most analysts would say "Time to trade Dragic!" and hit the trade machine for some fake trades. But how about we look into how we can make Dragic line-ups more effective....

30% of Dragic's shot attempts are from 3 vs the Heat's team average of 39% of shots from 3. One solution may be encouraging Dragic to change his mix of shots and shoot 10% more 3's (getting him in line with Miami's average) and ditch some drives to the basket that aren't generating free throws anyway.

This would mean getting up 1.5 more 3's per game (going from 4.1 to 5.6 attempts) and taking them away from drives.

Seems pretty reasonable in terms of applying the mix change, with Dragic shooting 36% from 3 this season which is a good rate. It wouldn't be a drastic change to what Dragic is doing, but will impact positively on line-ups he is a part of.

It's also much easier to adjust Dragic's game in terms of shot mix vs working on getting him to the free throw line more often at this stage of his career. Perhaps something for Dragic to work on in the off-season.

Again, I hope this post showed the power of what machine learning can do in terms of honing in on specific line-ups, then taking those findings and digging a little deeper into the numbers.

Ultimately "analytics" (I hate referring to it as that) is about taking all of this data and putting it into a format and language that coaches can make adjustments with, otherwise it's just useless numbers on a page.

Again I hope I have demonstrated how that can be done with the Dragic analysis.

Next up, I'll dive into the Cleveland Cavaliers.

(Thanks again to @nbastuffer https://www.nbastuffer.com/  for providing all of the NBA play by play data)