PLAYERS CLUSTERS
DEFINING ADVANCED ROLES IN BASKETBALL
Basketball has constantly evolved and players' characteristics have changed accordingly, by adapting their way of playing to a new style of basketball. Classifying players based on traditional roles (pointguard, shooting-guard, small-forward, power-forward and center) may be reductive and often don't help to understand the added value of a player to its team. The purpose of this analysis is to define a new classification scheme for players, based on physical and technical statistics recorded in the last Euroleague season (2020-2021).
CLUSTERING ALGORITHM
In order to define the players' roles, we use the K-means clustering algorithm. K-means is an unsupervised clustering algorithm that identifies groups of similar players by using the Euclidean distance between observations of a data set. The number of selected clusters to identify the new roles is 10.
DATA AND CLUSTERING PROCESS
Each cluster defines a new role.
 |
| *All statistics are per 36 minutes because the role classification should not be affected by the amount of playing time |
CLUSTER 1 - LOW USAGE STRETCH WINGS
In the first cluster we can find players with average size who can shoot from the distance but they don't excel in either one of the above listed statistical categories.
Player prototype: James Anderson, Alex Abrines, Evgeney Valiev, Tadas Sedekerskis, Andrey Zubkov, Angelo Caloiaro, Sergey Karasev, Janis Strelnieks, Marko Simonovic, Kostas Papanikolaou
CLUSTER 2 - ALL AROUND SCORING WINGS
Players in this group have average weight and height and they can do everything on the basketball court. They are above average in many statistical categories such as: pts, true shooting percentage, free throws attempted, possessions and steals.
Player prototype: Shavon Shields, William Howard, Simone Fontecchio, Tornike Shengelia, Darrun Hilliard, Ognjen Dobric, David Lighty, Luke Sikma, Mateusz Ponitka, Krunoslav Simon
CLUSTER 3 - OFFENSE ENGINE GUARDS
This cluster is characterized by undersized players who are used to playing a large number of possessions, who score many points for their teams, and with high true shooting percentage, free throw attempts, assists, and turnover. We can define these players as being the stars of their teams; indeed teams build their rosters around this type of players to maximize their skills.
Player prototype: Jordan Loyd, Nemanja Nedovic, Alexey Shved, Iffe Lundberg, Wade Baldwin, Nando De Colo, Vasilije Micic, Scottie Wilbekin, Shane Larkin, Sergio Rodriguez
CLUSTER 4 - TRADITIONAL BIGS
Oversized players (height & weight) characterized by a large number for blocks, rebounds, points and high true shooting percentage.
Player prototype: Tibor Pleiss, Ben Lammers, Ante Zizic, Edy Tavares, Nikola Milutinov, George Papagiannis, Moustapha Fall
CLUSTER 5 - HIGH USAGE SCORING BIGS
Bigs, points, true shooting percentage, many possessions and free throw attempts.
Player prototype: Jalen Reynolds, Nikola Mirotic, Youssupha Fall, Brandon Davies, Sertac Sanli, Jordan Mickey, Arturas Gudaitis
CLUSTER 6 - ATTACK MINDED GUARDS
Small players with many assists, high turnover, possessions, three point attempts, and steals. Players in this cluster are similar to the "offense engine guards" but they don't score as much as the offense guards do and they are not as efficient as the others.
Player prototype: Lukas Lekavicius, Lorenzo Brown, Chris Jones, Kostas Sloukas, Pierria Henry, Nicolas Laprovittola, Jayson Granger, Nick Calathes, Thomas Walkup, Antoine Diot
CLUSTER 7 - FLOOR SPACING WINGS (STRETCH 4)
3pt specialist wings that are also above average concerning rebounders.
Player prototype: Anthony Randolph, Alec Peters, Trey Thompkins, Georgios Printezis, Kostas Mitoglou, Vladimir Lucic, Guershon Yabusele, Luigi Datome, Paul Zipser, Derrick Williams, Achille Polonara
CLUSTER 8 - DEFENSIVE BALL HANDLER
Undersized players with many steals, fouls, assists, and possessions.
Player prototype: John DiBartolomeo, Adam Hanga, Nick Weiler-Babb, Vyacheslav Zaytsev, Leo Westermann, Lefteris Bochoridis, Giannoulis Larentzakis
CLUSTER 9 - BIG & ATHLETIC DEFENDERS
BIG (height,weight) players with many blocks, rebound, and fouls. They are similar to "traditional bigs" in cluster4 but they are less productive,less effective and they are prone to make many fouls.
Player prototype: Joffrey Lauvergne, Alex Poythress, Augustine Rubit ,Bryant Dunston, Othello Hunter, Livio Jean-Charles, Pierre Oriola, Landry Nnoko, Kyle Hines, Kaleb Tarczewski
CLUSTER 10 - SHOOTING GUARDS
Lots of three points attempted,above average concerning true shooting percentage.
Player prototype: Jaycee Carroll, Marcus Eriksson, Billy Baron, Rodrigue Beaubois, Kyle Kuric, Rokas Giedraitis, Vanja Marinkovic, Rudy Fernandez, Austin Hollins, Melih Mahmutoglu
FRONTCOURT AND BACKCOURT IN ADVANCED ROLES
EVALUATING ROLE EFFICIENCY BASED ON PER
PER = Player Efficiency Rating
"The PER sums up all a player's positive accomplishments, subtracts the negative accomplishments, and returns a per-minute rating of a player's performance"
Def BasketballReference
According to PER the most efficient role is high usage scoring bigs followed by traditional bigs. This is not surprising because PER is a statistic biased towards frontcourt players since they shoot closer to the basket compared to backcourt players, they grab a lot of rebounds and they rarely have to create for their teammates and risk turnovers.
PER is a statistic biased towards big players, this is why it make sense to look at player efficiency according to frontcourt and backcourt roles.
EVALUATING ROLE EFFICIENCY BASED ON DEFENSIVE RATING
Firstly, it is important to remark that defensive rating is not an individual stat.
Defensive rating = points allowed per 100 possessions by team X when a player Y is on the court
This table shows that big players on the court are the key roles for having a good defense. High usage scoring bigs affects team defense more than any other role. Defensive ball-handler is the only backcourt role that has a positive defensive impact.
EVALUATING ROLE EFFICIENCY BASED ON OFFENSIVE RATING
Offensive rating = points scored per 100 possessions by player Y is on the court
The key roles for the offensive side of the floor are still big players like traditional bigs and high usage scoring bigs.
FINAL RESULTS
Big players have a significant effect on their teams on both sides of the court. Looking at the offensive rating, we can notice that the roles that are able to shoot threes points at high percentages tend to have a high offensive rating like shooting guards and floor spacing wings.
On the other hand, "shooters specialists" perform low on the defensive part of the game, as you can see in the defensive rating table.Therefore, it becomes a difficult task for coaches and their staffs to find a good balance between an optimal usage of "shooting specialists" during the offensive stage of the game and their vulnerability and low productivity in the defensive stage.
TEAMS CLUSTERS
TEAMS CLUSTERING BASED ON ADVANCE ROLES
The purpose of this analysis is to classify teams based on the advanced roles to find the teams with similar roster construction and find the relation between the teams clusters and the final standings of the last year.
CLUSTERING ALGORITHM
In order to assess the teams with similar roster construction we use K-means clustering. K-means is an unsupervised clustering algorithm that identifies groups of similar teams by using Euclidean distance between observations of a data set. The number of selected clusters that represent the new roles is 4.
DATA AND CLUSTERING PROCESS
 |
| PLAYERS - The 8 players with the most minutes per game for each team. |
CLUSTER BLUE
CLUSTER TEAMS: Cska, Milano, Maccabi, Zvezda Belgrade
CLUSTER RANKING: Average standings position 8.75
CLUSTER STRENGTHS: Offense engine guards, all around scoring wings, traditional bigs, big athletic defenders
CLUSTER ROSTER CONSTRUCTION:
- more than 1 offensive engine guards to coordinate the offense helped by all around scoring wings
- the defensive side is covered by traditional bigs and big athletic defenders
- resume: backcourt create offense, frontcourt provide paint protection
CLUSTER YELLOW
CLUSTER TEAMS: Berlin, Asvel, Panathinaikos, Real Madrid
CLUSTER RANKING: Average standings posItion 12.8
CLUSTER STRENGTHS: Traditional bigs, all around scoring wings, defensive ball handlers.
CLUSTER ROSTER CONSTRUCTION:
- Frontcourt roles have no other options except for traditional bigs.
- Players with high offensive production like shooting guards and all around scoring wings in the backcourt roles but few players that can score and create for their teammates like offense engine guards and attack minded guards
CLUSTER RED
CLUSTER TEAMS: Baskonia,Olympiakos,Valencia,Zalgiris
CLUSTER RANKING: Average standings position 10.5
CLUSTER STRENGTHS: Big athletic defenders, floor spacing wings, shooting guards, attacked minded guards.
CLUSTER ROSTER CONSTRUCTION:
- Solid teams with high score in attack minded guards and shooting specialist roles like shooting guards and floor spacing wings.
- The gap between those teams and the best teams in the Euroleague is the lack of Traditional bigs, High usage scoring bigs and Offense engine guards that as we saw in the previous paragraphs are the most efficient roles according to PER.
CLUSTER GREEN
CLUSTER TEAMS: Efes, Barcelona, Fenerbahce, Khimki, Zenit, Bayern
CLUSTER RANKING: Average standings position 7.17
CLUSTER STRENGTHS: Teams of this cluster are complete in every role. Full of players such as floor spacing wings, high usage scoring bigs, offense engine guards and defensive ball handlers. Above average in the remaining roles.
CLUSTER ROSTER CONSTRUCTION:
- offense production provided by a combination of high usage scoring bigs and offense engine guards
- surrounded by shooters (shooting guards and floor spacing wings)
- presence of many defensive ball handlers, in fact this group is devoted to perimeter defense rather than paint defense (few player for the paint protection like big athletic defenders and traditional bigs)
TEAMS CLUSTER SHOOTING CHART AND DEFENSIVE SHOOTING CHART
 |
| Cluster Blue |
 | |
|
 |
| Cluster Red |
 |
| Cluster Green |
THE "HIGH RANKING" CLUSTERS AND "LOW RANKING" CLUSTERS COMPARISON
In fact the 22 offense engine guards listed by the algorithm the majority belong to Playoff teams and Final4 teams:
- 13/22 players (59%) belongs to Playoff teams
- 8/22 players (36%) belongs to Final4 teams
Quick reminder of offense engine guards: Micic, Larkin, Decolo, Shved, Punter, Pangos....
FINAL RESULTS
The backcourt role (offense engine guards) and the frontcourt role (high usage scoring bigs) with the higher PER have an important impact for their teams, so we analyze the minutes percentage of the total made by these roles for each team and then we search a relation between this stat and the winning percentage.
The result:
- 7/10 of teams (70%) over 20% of minutes made by offense engine guards or high usage scoring bigs have reached the Playoff.
- 3/4 of teams (75%) over 29% of minutes made by offense engine guards or high usage scoring bigs have reached the Final Four.
AUTHOR: Tommaso Del Prete
Comments
Post a Comment