The Positionless NBA: Grouping players using performance stats

By Josh Margles

Nikola Jokic is one of the most unique players in the NBA today. He is 7 feet tall but can pass like a point guard, shoot the three and has arguably the best vision among all players regardless of position, however, he is listed as a centre purely based on his height. He is redefining how the league classifies players because as a centre he averaged 7.3 assists per game and brings up the ball regularly for the Nuggets. Basketball is becoming more positionless with each season, so it is a wonder why the NBA continues to segment players with the outdated 5 positions. Instead of putting players into positions based on height and traditional labels, it would be more beneficial to sort the players based on their stats. Just because a player like Jokic is 7 feet, if he is the one taking the ball up and directing the offense, he is the point guard of the team.

To segment players into groups, I used k means clustering to put the players into segments based on their performance in the 2018-19 season. The k means algorithm attempts to find clusters of players based on their stats and puts the players into a group with similar players. The algorithm uses the smallest distance between the players stats and the cluster’ stats to find which group the player belongs in. Instead of arbitrarily putting a player into a category based on height, he is put into a cluster with players of a similar calibre.

For this project, the categories used for evaluating the players are points, assists, rebounds, field goal percentage, free throw percentage, three point makes, blocks and steals all on a per game basis. Since the NBA uses five positions, I had the model break the top 68 players of the 2018-19 season (based on minutes and games played) into five clusters. Here is a breakdown of the median stats for each of the groups:

\"Screen

Now let’s look at the players in each cluster. In cluster 1, the group is filled with a lot of traditional point guards like Chris Paul, Trae Young and Mike Conley. As shown above, this cluster has the highest assist per game and doesn’t rebound that well. Also, in this group is both Lebron James and Jokic. While they aren’t considered traditional point guards, their assist numbers along with their average amount of three point makes puts them with a lot of other guards. Two more interesting members are Demar Derozan and Jimmy Butler. Neither of their assist numbers are high but their rebounds, blocks and points per game numbers are in line with the cluster.

The second cluster as shown above is the last in terms of points per game, FT%, assists and threes made, while being first in FG% and second in rebounds. This group is full of low scoring, high efficiency big men like Rudy Gobert, DeAndre Ayton and Clint Capela. The most interesting name on the list is Ben Simmons. He is classified as a point guard but other than his assist numbers he does everything else like a big. He doesn’t shoot threes, rebounds close to 9 per game and has both a high field goal rate with a low free throw rate. He is another player redefining the traditional positions with is unique skillset and stats.

Cluster 3 is the smallest group but has plenty of big names. The cluster rebounds the highest and is a close second in scoring. The group is efficient, can shoot the three and block a shot. The cluster has some of the best big men like Anthony Davis, Joel Embiid and Karl-Anthony Towns and two players in Kevin Durant and Giannis who rebound well and play with the efficiency of an elite big. Some might be surprised to see LaMarcus Aldridge and Nikola Vucevic in this group, but it just shows how well they did last season. They are the lowest scorers of this group, but both shot 52% from the field, gathered over 9 boards, blocked over 1 shot a game while averaging 20+ points. Vucevic is particularly impressive averaging 21 and 12 along with a block, a steal and a three per game (the only player to do so this season).

Cluster 4 is by far the largest and is mostly shooting guards and forwards. The algorithm puts these players together because almost none of these players excel at more than one or two stats. There are some impressive players here like all stars Blake Griffin, Klay Thompson and Khris Middleton but none of them are good enough to separate from the pack. No one in the cluster averaged more than 8 rebounds, 6 assists or 25 points per game.

The final cluster is packed with scorers. Steph Curry, James Harden, Devin Booker and Paul George all were top 10 scorers last year and shoot a lot of threes. Everyone in this cluster averaged more than 23 points per game last season. This cluster also is all guards except for Kawhi Leonard and George who are classified as forwards. It is hard to find a cluster for Kawhi as he is so unique in mixing high scoring and defense, with shooting efficiency with a low assist total. The reason why he is put in this group is that he scores well and steals the ball like a guard but out of his cluster has the fewest threes made (and attempts). Kawhi is another player that does not really have a position as he can rebound like a big man, shoot like a guard and guard multiple positions on the defensive end.

\"IMG_7634.jpeg\"

Leave a Comment

Your email address will not be published. Required fields are marked *