NHL Machine Learning Draft

Happy Father's day! (tomorrow).

As noted in the 2019 sleepers post, the machine learning model that's used to predict a player's career performance provides some intermediate data from other information data can be gleaned.

One such piece of this intermediate data can loosely be interpreted as the probability that the given player becomes a superstar. Note that ranking prospects by 'superstar power' might be misleading in some cases. For example, we can have player A who has a 50% chance of becoming a superstar, but a 50% chance of missing the NHL, and player B, who only has a 60% chance of being a superstar, but has a 40% chance of being a 2nd liner, say. So, while not all-inclusive metric, this 'superstar probability' might be of interest for the 'swinging for the fences' picks.

Let's look at the top-9 players as ranked by their superstar probability. Remember the model determines it's unable to make a prediction for some players. See the 2019 predictions; If a player is missing there, he won't show up here either.

Okay, let's look at some players:

9. Nathan Legare -- Superstar probability: 15% Legare was mentioned in the 2019 sleepers article. Here's what we said about him there:

(Sorry about missing accents. That's a site-wide issue that needs to be addressed at some point).

Legare had a superb season in the QMJHL with Baie-Comeau (45G, 42A, in 68GP). He gained 12 spots in the NA CSS rankings between miderm and final (was 64th, now 52nd).

Legare's age is a bit of a mitigating factor -- he's on the older side of the 2019 draft cohort. At 200 lbs., he also doesn't have the 'headroom' for more healthy weight that some prospects may have.

The ML model predicts he'd be an excellent value for a 3rd round pick. We'll see if he goes earlier in the draft or if someone ends up with what would be a 3rd round steal.

8. Trevor Zegras -- Superstar probability: 47% In terms of goals and assists, Zegras might be considered the ying to Cole Caufield's (see below) yang. Zegras put up 61 assists and 26 goals in 60 games with the US National squad. Mock drafts suggest he'll be drafted somewhere between 6th and 10th in the 2019 NHL draft. Insider note: the majority of the machine learning models seem to like players who pile up assists.

7. Cole Caufield -- Superstar probability: 49% Perhaps one of the most talked-about prospects available in the 2019 NHL draft, Caufield's lack of size (just 5'7") is contrasted by his gaudy statline with the US National Team Development Program (72G, 28A in 64 GP). The buzz is warranted as Caufield feels like a real boom or bust pick. Nobody wants to whiff in the top 10, but nobody wants to be left thinking 'what if', either. It will be interesting to see where he goes on draft day.

6. Bobby Brink -- Superstar probability: 52% Another undersized player at 5'8", Brink scored 35 goals and added 33 assists in 43 games with the Sioux City Muskateers of the USHL. The fact that he's on the younger side of the 2019 NHL draft cohort (turns 18 on July 8, 2019) adds to the intrigue.

5. Kaapo Kakko -- Superstar probability: 58% Kakko is a well-known prospect at this point; I won't add anything.

A ranking of five on this list seems a bit low, but as mentioned above, this isn't an an all-encompasing prediction of a prospect's success.

4. Matthew Boldy -- Superstar probability: 61% Boldy's got a big frame (6'2", 192 lbs.), and put up solid numbers with the NTDP (33G, 48A in 66 GP). Buzz wise, seems to be flying a bit under the radar amidst all the discussion about Hughes, Caufield, and Zegras.

3. Jack Hughes -- Superstar probability: 61% Like Kakko, there's not much more to be said about Hughes that hasn't been said already.

2. Alex Turcotte -- Superstar probability: 75% Turcotte appears to have carved out some space as a tier "1.5" forward after his value rose dramatically between mid term and season's end. Another of the eight or so US National Team Development Players slated to go in the first round,

1. Arthur Kaliyev -- Superstar probability: 78% A bit of a surprise to see at number one, opinions seem to vary widely on Kaliyev. Ranked as low as #34 by some scouting services, the Central Scouting service has him at #7. Kaliyev had a huge season in terms of production with the Hamilton Bulldogs of the OHL (51G, 51A in 67G). On the other hand, he had a +/- of -13, and a +/- of -8 in just 4 playoff games.

All told, the machine learning model has probability over-estimated Kaliyev's superstar probability, but he still looks like a high-risk, high-reward type of pick.

Who's Missing?

The most notable absences from the above list would probably be Kirby Dach (15th on the this list), Dylan Cozens (11th on the this list), and Peyton Krebs (25th on the this list). This group seems to be part of the tier 2 forward group, with many mock drafts picking one or more of the above in the top 10. It's worth noting that the current 2019 predictions still think highly of Cozens despite not showing well in the metric under discussion.

What do you think? Let me know on twitter.