Anyone who has analyzed historical NBA draft picks should know that translating college success to the NBA is never a sure thing. No matter how good the scout, GM, or draft model, there is still a high level of uncertainty involved in predicting NBA success.
Therefore, my goal with this year's draft model is to highlight this uncertainty. By putting less emphasis on a single number, and more emphasis on the range of outcomes, our evaluation of players can be more representative of the roll of the dice that NBA teams actually face.
If you just want results, jump ahead to the 2018 rankings!
Modeling Details
Data
The models described below are based on custom statistics derived from player game log data1 while controlling for efficiency (minutes), level of competition (opponent), and situation (team). These are viewed as enhanced variations of pace and minute adjusted numbers, as they have an ability to better capture a players ability in situations where a coach (defensive system) or opponent (playing style) might be impacting statistical performance. Additionally, by factoring in individual game performances, strong (or weak) performances against the best competition are not lost in season averages. Other inputs also include scouting (high school ranking) and player characteristics (age, height, weight, combine).
Quant Model
Quant is a gradient boosting quantile regression model which predicts a players NBA peak single season win shares. By using a quantile regression model, it allows you to evaluate predictions across quantiles, highlighting differences in risk and upside.
By looking at the 25th, 50th, 75th, and 90th percentile outcomes for a player we can see that a player like Markelle Fultz might have a slightly higher upside (q90) than Jayson Tatum, but a slightly lower median expected outcome (q50).
APEX Model
While I recommend focusing on the full spectrum of Quant outcomes, I realize there are times when a single number is unavoidable (i.e. draft board rank). Therefore, APEX is a single number interpretation of Quant. Given that star players are so important in the NBA, APEX is calculated by averaging the higher upside outcomes (q75 and q90). In cases where a player has multiple season's worth of pre-NBA data, the players overall APEX value is a minute weighted average of all APEX values.
Below are the top 25 players as rated by APEX.
While there is room for improvement, historical (out of sample) results below show that, in general, performance is quite good.
Star Model
Star is a classification model which predicts the probability of being an All Star (star), Starter (start), Role Player (role), or Bust (bust).2 The advantage of the Star model is that the results are intuitive, however keep in mind that being a Starter or Role Player in the NBA is sometimes about situation and does not imply success or efficiency (which is what is captured in the Quant model).
Below you can see that Greg Oden had a wide range of outcomes with a higher probability of being either a bust or a star, whereas Kevin Durant had a more narrow range of outcomes and a smaller probability of becoming a bust.
Comps
To give insight into playing style and NBA combine comparables, I have a K-nearest neighbor model which shows historical players that are most similar. Below we can see that Trae Young and Steph Curry are indeed similar, however they do have some key differences in assists and steals.
2018 Rankings
Top 60
Below you can see the top 60 APEX rated players and the complete Top 100 can be found here.
Top Guards
Below is a heat map showing a players strengths and weaknesses which can help you understand why the models might view a player a particular way. See the glossary for any non-obvious column names. The complete list of Guards in the Top 100 can be found here.
Top Bigs
You can find the complete Top 100 here.
All Star Potential
Below are the players with the highest NBA All Star potential probability.
Notice how Shai has a high probability of being a solid starter, however a lower All Star potential compared to some other risker players.
Mock Comparison
Here is how APEX ranks players in comparison with various draft boards (all updated on 6/18):
Contact me on Twitter if you think another ranking should be included. The table includes players who appeared in the Top 60 on any of the draft boards.
Player Profiles
Below you can find lots of information about top players including:
- Statistical Profile - visualize the strengths, weaknesses, and growth trajectory of individual players over their college careers
- Playing Style Comps - find the most similar player seasons compared to the potential draftees most recent season, which can keep you honest on a players expected outcome
- Combine Comps - find the most similar players in terms of physique and athleticism
- Draft Model Likes/Dislikes - gain a better understanding as to why my draft model might be ranking a player higher/lower than expected3
1. Wendell Carter Jr. - Duke
FR | F | 6-8 | 251 | rsci: 7 | s-ref | realgm
Statistical Profile
Playing Style Comps
Combine Comps
Draft Model Likes
- age
- drb rate
- height
Draft Model Dislikes
- 3p rate
- scoring rate
- stl rate
2. Jaren Jackson Jr. - Michigan State
FR | F | 6-9 | 236 | rsci: 9 | s-ref | realgm
Statistical Profile
Playing Style Comps
Combine Comps
Draft Model Likes
- age
- shooting efficiency
- 2p shooting
Draft Model Dislikes
- 3p rate
- stl rate
- ast rate
3. Trae Young - Oklahoma
FR | G | 6-0 | 177 | rsci: 21 | s-ref | realgm
Statistical Profile
Playing Style Comps
Combine Comps
Draft Model Likes
- scoring rate
- age
- drb rate
Draft Model Dislikes
- stl rate
- shooting efficiency
- 2p shooting
4. Luka Doncic - Real Madrid
G | 6-6 | 225 | s-ref | realgm
Statistical Profile
Playing Style Comps
Combine Comps
Draft Model Likes
- age
- 2p shooting
- ast rate
Draft Model Dislikes
- 3p shooting
- scoring rate
- stl rate
5. DeAndre Ayton - Arizona
FR | F | 6-11 | 247 | rsci: 3 | s-ref | realgm
Statistical Profile
Playing Style Comps
Combine Comps
Draft Model Likes
- age
- drb rate
- scoring rate
Draft Model Dislikes
- stl rate
- ft rate
- ft shooting
6. Shai Gilgeous-Alexander - Kentucky
FR | G | 6-4 | 180 | rsci: 30 | s-ref | realgm
Statistical Profile
Playing Style Comps
Combine Comps
Draft Model Likes
- stl rate
- age
- weight
Draft Model Dislikes
- drb rate
- ft shooting
- shooting efficiency
7. Troy Brown - Oregon
FR | F | 6-5 | 208 | rsci: 12 | s-ref | realgm
Statistical Profile
Playing Style Comps
Combine Comps
Draft Model Likes
- age
- stl rate
- ast rate
Draft Model Dislikes
- ft shooting
- shooting efficiency
- scoring rate
8. Dzanan Musa - Cedevita
G | 6-7 | 192 | s-ref | realgm
Statistical Profile
Playing Style Comps
Combine Comps
Draft Model Likes
- age
- height
- weight
Draft Model Dislikes
- ft shooting
- shooting efficiency
- drb rate
9. Marvin Bagley III - Duke
FR | F | 6-9 | 231 | rsci: 1 | s-ref | realgm
Statistical Profile
Playing Style Comps
Combine Comps
Draft Model Likes
- age
- shooting efficiency
- 2p shooting
Draft Model Dislikes
- ast rate
- drb rate
- stl rate
10. Miles Bridges - Michigan State
SO | G | 6-5 | 220 | rsci: 10 | s-ref | realgm
Statistical Profile
Playing Style Comps
Combine Comps
Draft Model Likes
- age
- ast rate
- shooting efficiency
Draft Model Dislikes
- scoring rate
- stl rate
- ft rate
11. Robert Williams - Texas A&M
SO | F | 6-7 | 234 | rsci: 51 | s-ref | realgm
Statistical Profile
Playing Style Comps
Combine Comps
Draft Model Likes
- ast rate
- drb rate
- 2p shooting
Draft Model Dislikes
- weight
- ft rate
- age
12. De'Anthony Melton - USC
FR | G | 6-2 | 193 | s-ref | realgm
Statistical Profile
Playing Style Comps
Combine Comps
Draft Model Likes
- age
- stl rate
- drb rate
Draft Model Dislikes
- height
- ft shooting
- scoring rate
13. Mohamed Bamba - Texas
FR | F | 6-11 | 225 | rsci: 4 | s-ref | realgm
Statistical Profile
Playing Style Comps
Combine Comps
Draft Model Likes
- age
- drb rate
- shooting efficiency
Draft Model Dislikes
- ft shooting
- ft rate
- ast rate
14. Rodions Kurucs - FC Barcelona Lassa
F | 6-8 | 187 | s-ref | realgm
Statistical Profile
Playing Style Comps
Combine Comps
Draft Model Likes
- 2p shooting
- age
- ast rate
Draft Model Dislikes
- ft shooting
- ft rate
- drb rate
15. Kevin Huerter - Maryland
SO | G | 6-6 | 194 | rsci: 53 | s-ref | realgm
Statistical Profile
Playing Style Comps
Combine Comps
Draft Model Likes
- 2p shooting
- shooting efficiency
- age
Draft Model Dislikes
- stl rate
- scoring rate
- drb rate
16. Zhaire Smith - Texas Tech
FR | G | 6-2 | 198 | s-ref | realgm
Statistical Profile
Playing Style Comps
Combine Comps
Draft Model Likes
- age
- shooting efficiency
- 2p shooting
Draft Model Dislikes
- ft shooting
- height
- drb rate
17. Mikal Bridges - Villanova
JR | G | 6-5 | 188 | rsci: 96 | s-ref | realgm
Statistical Profile
Playing Style Comps
Combine Comps
Draft Model Likes
- stl rate
- shooting efficiency
- 2p shooting
Draft Model Dislikes
- tov rate
- height
- age
18. Jarred Vanderbilt - Kentucky
FR | F | 6-7 | 211 | rsci: 13 | s-ref | realgm
Statistical Profile
Playing Style Comps
Combine Comps
Draft Model Likes
- age
- ast rate
- drb rate
Draft Model Dislikes
- ft shooting
- shooting efficiency
- 2p shooting
19. Brandon McCoy - UNLV
FR | F | 6-11 | 250 | rsci: 11 | s-ref | realgm
Statistical Profile
Playing Style Comps
Combine Comps
Draft Model Likes
- age
- scoring rate
- 3p shooting
Draft Model Dislikes
- 2p shooting
- ast rate
- shooting efficiency
20. Moritz Wagner - Michigan
JR | F | 6-10 | 241 | rsci: 97 | s-ref | realgm
Statistical Profile
Playing Style Comps
Combine Comps
Draft Model Likes
- drb rate
- stl rate
- height
Draft Model Dislikes
- tov rate
- 3p shooting
- ast rate
21. Chandler Hutchison - Boise State
SR | G | 6-5 | 187 | s-ref | realgm
Statistical Profile
Playing Style Comps
Combine Comps
Draft Model Likes
- drb rate
- ast rate
- ft rate
Draft Model Dislikes
- 2p shooting
- shooting efficiency
- age
22. Josh Okogie - Georgia Tech
SO | G | 6-3 | 210 | s-ref | realgm
Statistical Profile
Playing Style Comps
Combine Comps
Draft Model Likes
- age
- scoring rate
- stl rate
Draft Model Dislikes
- height
- shooting efficiency
- 2p shooting
23. Landry Shamet - Wichita State
JR | G | 6-4 | 188 | s-ref | realgm
Statistical Profile
Playing Style Comps
Combine Comps
Draft Model Likes
- 2p shooting
- shooting efficiency
- ast rate
Draft Model Dislikes
- age
- stl rate
- scoring rate
24. Elie Okobo - Pau-Lacq-Orthez
G | 6-1 | 177 | s-ref | realgm
Statistical Profile
Playing Style Comps
Combine Comps
Draft Model Likes
- shooting efficiency
- 2p shooting
- scoring rate
Draft Model Dislikes
- height
- weight
- stl rate
25. Kevin Hervey - Texas-Arlington
SR | F | 6-6 | 211 | s-ref | realgm
Statistical Profile
Playing Style Comps
Combine Comps
Draft Model Likes
- drb rate
- ft shooting
- height
Draft Model Dislikes
- scoring rate
- 2p shooting
- age
26. Kevin Knox - Kentucky
FR | F | 6-7 | 212 | rsci: 10 | s-ref | realgm
Statistical Profile
Playing Style Comps
Combine Comps
Draft Model Likes
- age
- height
- weight
Draft Model Dislikes
- stl rate
- scoring rate
- ast rate
27. Collin Sexton - Alabama
FR | G | 6-0 | 183 | rsci: 6 | s-ref | realgm
Statistical Profile
Playing Style Comps
Combine Comps
Draft Model Likes
- age
- scoring rate
- 3p shooting
Draft Model Dislikes
- height
- drb rate
- stl rate
28. Michael Porter Jr. - Missouri
FR | F | 6-9 | 211 | rsci: 2 | s-ref | realgm
Statistical Profile
Playing Style Comps
Combine Comps
Draft Model Likes
- stl rate
- age
- height
Draft Model Dislikes
- ast rate
- 2p shooting
- shooting efficiency
29. Khyri Thomas - Creighton
JR | G | 6-2 | 198 | s-ref | realgm
Statistical Profile
Playing Style Comps
Combine Comps
Draft Model Likes
- 2p shooting
- shooting efficiency
- stl rate
Draft Model Dislikes
- ft rate
- drb rate
- age
30. Williams Narace - Nancy
F | 6-6 | 202 | s-ref | realgm
Statistical Profile
Playing Style Comps
Combine Comps
Draft Model Likes
- stl rate
- age
- 2p shooting
Draft Model Dislikes
- ft shooting
- height
- ast rate
31. Jerome Robinson - Boston College
JR | G | 6-4 | 188 | s-ref | realgm
Statistical Profile
Playing Style Comps
Combine Comps
Draft Model Likes
- 2p shooting
- shooting efficiency
- ast rate
Draft Model Dislikes
- age
- drb rate
- stl rate
32. Chimezie Metu - USC
JR | F | 6-8 | 219 | rsci: 35 | s-ref | realgm
Statistical Profile
Playing Style Comps
Combine Comps
Draft Model Likes
- height
- shooting efficiency
- 2p shooting
Draft Model Dislikes
- ast rate
- stl rate
- drb rate
33. Alize Johnson - Missouri State
SO | F | 6-7 | 216 | s-ref | realgm
Statistical Profile
Playing Style Comps
Combine Comps
Draft Model Likes
- drb rate
- ast rate
- weight
Draft Model Dislikes
- ft rate
- stl rate
- age
34. Melvin Frazier - Tulane
JR | G | 6-4 | 198 | s-ref | realgm
Statistical Profile
Playing Style Comps
Combine Comps
Draft Model Likes
- shooting efficiency
- 2p shooting
- stl rate
Draft Model Dislikes
- ft rate
- scoring rate
- age
35. Jacob Evans - Cincinnati
JR | G | 6-4 | 199 | s-ref | realgm
Statistical Profile
Playing Style Comps
Combine Comps
Draft Model Likes
- ft rate
- stl rate
- 3p rate
Draft Model Dislikes
- height
- scoring rate
- age
36. Keenan Evans - Texas Tech
SR | G | 6-1 | 183 | s-ref | realgm
Statistical Profile
Playing Style Comps
Combine Comps
Draft Model Likes
- shooting efficiency
- 2p shooting
- scoring rate
Draft Model Dislikes
- drb rate
- stl rate
- age
37. Tryggvi Hlinason - Valencia Basket
F | 6-11 | 252 | s-ref | realgm
Statistical Profile
Playing Style Comps
Combine Comps
Draft Model Likes
- scoring rate
- stl rate
- height
Draft Model Dislikes
- weight
- ft shooting
- drb rate
38. Rawle Alkins - Arizona
SO | G | 6-2 | 217 | rsci: 21 | s-ref | realgm
Statistical Profile
Playing Style Comps
Combine Comps
Draft Model Likes
- stl rate
- ft rate
- 3p rate
Draft Model Dislikes
- drb rate
- 2p shooting
- shooting efficiency
39. Raymond Spalding - Louisville
JR | F | 6-9 | 215 | rsci: 41 | s-ref | realgm
Statistical Profile
Playing Style Comps
Combine Comps
Draft Model Likes
- stl rate
- height
- weight
Draft Model Dislikes
- 2p shooting
- drb rate
- age
40. Sviatoslav Mykhailiuk - Kansas
SR | G | 6-6 | 211 | rsci: 95 | s-ref | realgm
Statistical Profile
Playing Style Comps
Combine Comps
Draft Model Likes
- stl rate
- ast rate
- height
Draft Model Dislikes
- ft shooting
- drb rate
- shooting efficiency
41. Keita Bates-Diop - Ohio State
SR | F | 6-7 | 223 | rsci: 29 | s-ref | realgm
Statistical Profile
Playing Style Comps
Combine Comps
Draft Model Likes
- drb rate
- ast rate
- ft shooting
Draft Model Dislikes
- weight
- stl rate
- age
42. Trevon Duval - Duke
FR | G | 6-1 | 191 | rsci: 5 | s-ref | realgm
Statistical Profile
Playing Style Comps
Combine Comps
Draft Model Likes
- stl rate
- age
- ast rate
Draft Model Dislikes
- ft shooting
- shooting efficiency
- drb rate
43. Jalen Brunson - Villanova
JR | G | 6-1 | 198 | rsci: 19 | s-ref | realgm
Statistical Profile
Playing Style Comps
Combine Comps
Draft Model Likes
- shooting efficiency
- 2p shooting
- ft shooting
Draft Model Dislikes
- weight
- age
- stl rate
44. Grayson Allen - Duke
SR | G | 6-3 | 198 | rsci: 24 | s-ref | realgm
Statistical Profile
Playing Style Comps
Combine Comps
Draft Model Likes
- stl rate
- ft rate
- 2p shooting
Draft Model Dislikes
- drb rate
- scoring rate
- age
45. Yante Maten - Georgia
SR | F | 6-7 | 246 | s-ref | realgm
Statistical Profile
Playing Style Comps
Combine Comps
Draft Model Likes
- scoring rate
- ast rate
- ft shooting
Draft Model Dislikes
- usg rate
- age
- stl rate
46. Kenrich Williams - Texas Christian
JR | G | 6-6 | 200 | s-ref | realgm
Statistical Profile
Playing Style Comps
Combine Comps
Draft Model Likes
- stl rate
- ast rate
- height
Draft Model Dislikes
- scoring rate
- shooting efficiency
- age
47. Shake Milton - SMU
JR | G | 6-4 | 207 | rsci: 89 | s-ref | realgm
Statistical Profile
Playing Style Comps
Combine Comps
Draft Model Likes
- drb rate
- ft rate
- 3p shooting
Draft Model Dislikes
- shooting efficiency
- age
- 2p shooting
48. Donte DiVincenzo - Villanova
JR | G | 6-3 | 200 | s-ref | realgm
Statistical Profile
Playing Style Comps
Combine Comps
Draft Model Likes
- shooting efficiency
- 2p shooting
- drb rate
Draft Model Dislikes
- height
- scoring rate
- age
49. Lonnie Walker - Miami (FL)
FR | G | 6-3 | 196 | rsci: 17 | s-ref | realgm
Statistical Profile
Playing Style Comps
Combine Comps
Draft Model Likes
- age
- ast rate
- weight
Draft Model Dislikes
- 2p shooting
- shooting efficiency
- drb rate
50. Malik Newman - Kansas
SO | G | 6-2 | 189 | rsci: 8 | s-ref | realgm
Statistical Profile
Playing Style Comps
Combine Comps
Draft Model Likes
- shooting efficiency
- drb rate
- ft shooting
Draft Model Dislikes
- 3p shooting
- height
- age
51. Thomas Welsh - UCLA
SR | C | 6-11 | 254 | rsci: 33 | s-ref | realgm
Statistical Profile
Playing Style Comps
Combine Comps
Draft Model Likes
- drb rate
- height
- stl rate
Draft Model Dislikes
- ft rate
- shooting efficiency
- age
52. Aaron Holiday - UCLA
JR | G | 5-11 | 187 | rsci: 60 | s-ref | realgm
Statistical Profile
Playing Style Comps
Combine Comps
Draft Model Likes
- weight
- ast rate
- height
Draft Model Dislikes
- ft shooting
- drb rate
- age
53. Theo Pinson - UNC
SR | F | 6-5 | 211 | rsci: 15 | s-ref | realgm
Statistical Profile
Playing Style Comps
Combine Comps
Draft Model Likes
- shooting efficiency
- 2p shooting
- drb rate
Draft Model Dislikes
- tov rate
- weight
- age
54. Bruce Brown - Miami (FL)
SO | G | 6-3 | 194 | rsci: 26 | s-ref | realgm
Statistical Profile
Playing Style Comps
Combine Comps
Draft Model Likes
- drb rate
- weight
- ft rate
Draft Model Dislikes
- 2p shooting
- scoring rate
- age
55. Allonzo Trier - Arizona
JR | G | 6-3 | 198 | rsci: 12 | s-ref | realgm
Statistical Profile
Playing Style Comps
Combine Comps
Draft Model Likes
- 2p shooting
- shooting efficiency
- ft shooting
Draft Model Dislikes
- height
- stl rate
- age
56. Jevon Carter - West Virginia
SR | G | 6-0 | 196 | s-ref | realgm
Statistical Profile
Playing Style Comps
Combine Comps
Draft Model Likes
- stl rate
- drb rate
- ft shooting
Draft Model Dislikes
- shooting efficiency
- 2p shooting
- age
57. Tony Carr - Penn State
SO | G | 6-3 | 198 | rsci: 43 | s-ref | realgm
Statistical Profile
Playing Style Comps
Combine Comps
Draft Model Likes
- ast rate
- age
- ft shooting
Draft Model Dislikes
- shooting efficiency
- stl rate
- 2p shooting
58. Isaac Haas - Purdue
SR | C | 7-1 | 303 | rsci: 82 | s-ref | realgm
Statistical Profile
Playing Style Comps
Combine Comps
Draft Model Likes
- 2p shooting
- shooting efficiency
- ft shooting
Draft Model Dislikes
- drb rate
- ast rate
- age
59. Devonte' Graham - Kansas
SR | G | 6-0 | 186 | rsci: 65 | s-ref | realgm
Statistical Profile
Playing Style Comps
Combine Comps
Draft Model Likes
- ast rate
- height
- tov rate
Draft Model Dislikes
- 2p shooting
- shooting efficiency
- age
60. Omari Spellman - Villanova
FR | F | 6-8 | 253 | rsci: 17 | s-ref | realgm
Statistical Profile
Playing Style Comps
Combine Comps
Draft Model Likes
- ft shooting
- drb rate
- 3p rate
Draft Model Dislikes
- shooting efficiency
- age
- weight
Glossary
- q25 = predicted 25th percentile outcome
- q50 = predicted 50th percentile outcome
- q75 = predicted 75th percentile outcome
- q90 = predicted 90th percentile outcome
- apex = average of q75 and q90
- rsci = high school rank
- max_ws = max single season win shares in the NBA
- star = NBA All Star probability
- start = NBA Starter probability (starter for >60 NBA games in a season)
- role = NBA Role Player probability (played in >60 NBA games in a season)
- bust = NBA Bust probability
- ht = Height
- wt = Weight
- gs = Games Started
- gp = Games Played
- mpg = Minutes Per Game
- ts = True Shooting Percentage
- efg = Effective Field Goal Percentage
- 3par = 3 Point Attempt Rate
- ftr = Free Throw Rate
- usg = Usage Rate
- bpm = Box Plus Minus
- srs = Team Quality (custom metric)
- sos = Strength of Schedule (custom metric)
- pk = NBA Draft Pick
- sim = Player Similarity
-
Older years do not have college game log data available, therefore the game log metrics are imputed. ↩
-
The definition of All Star is self-explanatory, whereas I arbitrarily defined Starter as having started 60 games in a NBA season, Role Player as having played 60 games in a NBA season, and Bust as everyone else. ↩
-
The like/dislike information is automatically generated by creating different models without certain pieces of data to see how a players prediction changes. It is not always perfect because of collinearity but I think I have got it tuned to do a pretty decent job at highlighting strengths and weaknesses. Let me know if you any results seem strange. ↩