Using machine learning to predict 2019 goals scored

Discussion in 'By The Numbers' started by Punn, Sep 14, 2018.

View Users: View Users
  1. Punn

    Punn Registered User

    Joined:
    Sep 24, 2011
    Messages:
    278
    Likes Received:
    3
    Trophy Points:
    36
    Location:
    Brampton, ON
    Hello all,

    Would appreciate if you check out my website (I put this together in a day).

    A built a ML model to predict goals for the 2019 season. If there's interested, I'll go through the CRISP on how I built the model :)

    Have a look and give some feedback :)
    2019 Goal Scoring Predictions
     
    Trilliann and Doctor No like this.
  2. LatvianTwist

    LatvianTwist Global Moderator

    Joined:
    Jul 23, 2010
    Messages:
    28,910
    Likes Received:
    952
    Trophy Points:
    165
    Totals in general seem pretty low across the board to me. Is there a particular reason for that?
     
    Doctor No likes this.
  3. Doctor No

    Doctor No Registered User

    Joined:
    Oct 26, 2005
    Messages:
    7,655
    Likes Received:
    864
    Trophy Points:
    139
    Home Page:
    I'd bet that it has to do with how he's removing outliers in his model (which he mentions but not in great detail) - two ways of doing this are to truncate and move on, or to truncate and spread, and from looking at the results, I'm betting on the former.

    The results are interesting, although I'd love to see additional details on the construction in order to properly evaluate the level of trust I give to the results (I do a fair amount of this in my day-to-day job, and have seen some great models and some poor ones).
     
  4. younggreybeard

    younggreybeard Registered User

    Joined:
    Sep 14, 2018
    Messages:
    1
    Likes Received:
    0
    Trophy Points:
    1
    Gender:
    Male
    Super interesting! Would also like to hear about how the model was made and what factors you included or left out!

    Also, did you test the model at all? i.e. Using 16-17 totals to predict 17-18 totals and then compare them against actual results?
     
  5. LeHab

    LeHab Registered User

    Joined:
    Aug 31, 2005
    Messages:
    11,612
    Likes Received:
    1,213
    Trophy Points:
    169
    He states to be using historical data but yeah no mention of how it was validated. I would also like to know which, if any, other attributes (age, team...) are used and what, if any, weighting is assigned.
     
    Last edited: Sep 15, 2018
  6. Punn

    Punn Registered User

    Joined:
    Sep 24, 2011
    Messages:
    278
    Likes Received:
    3
    Trophy Points:
    36
    Location:
    Brampton, ON
    Thanks everyone for the feedback.

    I will doing another post on the how:
    1. I defined the target,
    2. the features I used,
    3. a univariate analysis of all the features and the impact on target based on information gain,
    4. algorithm used including pre-processing, feature transformation (categorical and continuous variable), missing value imputation
    5. the lift from each feature and partial dependency,
    6. the most impactful features on the overall model,
    7. performance on the training, cross validation (5 CV), and holdout set based on RMSE and R-Squared
    Stay tuned and please subscribe to the website if you like what you see :)
     
  7. 2LeftSkates

    2LeftSkates Registered User

    Joined:
    Nov 20, 2008
    Messages:
    1,807
    Likes Received:
    253
    Trophy Points:
    94
    Algorithms do not shoot hockey pucks, people do. Video game voodoo.
     
  8. Bear of Bad News

    Bear of Bad News HFBoards Escape Goat

    Joined:
    Sep 27, 2005
    Messages:
    6,106
    Likes Received:
    1,390
    Trophy Points:
    139
    Location:
    Windsor
    You're welcome to participate in this sub-forum, but future responses such as this will be considered trolling and handled thusly.

    From the stickied thread at the top of the sub-forum:
    1. Respect what goes on here. If you can't do that, then feel free to ignore the forum entirely.
    2. Be patient, and trust others' motives. What's basic to you may not be basic to me, and we're all in this together. Be helpful. Take opportunities to learn, and take opportunities to teach.

    It's similar to (for instance) going into the Islanders forum and posting "Potvin sucks" in a thread, and you will be handled similarly. If your goal is to get attention, rest assured that you've got it.
     
  9. 2LeftSkates

    2LeftSkates Registered User

    Joined:
    Nov 20, 2008
    Messages:
    1,807
    Likes Received:
    253
    Trophy Points:
    94
    I wandered in by mistake. Not an attention hog, just saying what I thought. Don't worry, I won't be back in here(except maybe at the end of the year to see how it ends). No more posts, though.
     
    Bear of Bad News likes this.
  10. Bear of Bad News

    Bear of Bad News HFBoards Escape Goat

    Joined:
    Sep 27, 2005
    Messages:
    6,106
    Likes Received:
    1,390
    Trophy Points:
    139
    Location:
    Windsor
    Thank you.
     
  11. torniojaws

    torniojaws Registered User

    Joined:
    Jan 10, 2017
    Messages:
    529
    Likes Received:
    213
    Trophy Points:
    46
    Location:
    Finland
    Max Pacioretty (WPG)
    Did I miss a trade Vegas did? :)

    Does your model take into account goals "missed" due to injuries? Eg. Getzlaf 11 goals.
     

Share This Page

monitoring_string = "358c248ada348a047a4b9bb27a146148"