readme.md

NHL Game Classification - Game Win/Loss for 2023/24 Season

This project aims to classify NHL games as either a win or a loss by team, for the 2023/24 season.

Introduction

In this project, we will analyze historical NHL game data and build machine learning models (Classification) to predict the outcome of games for the 2023/24 season. The models will classify each game as either a win or a loss based on various features.

• Data is obtained using the NHL API

Run Locally

1.) Create and Activate the Virtual Environment

• Open a terminal and navigate to the points directory:

  cd .\examples\teams\

• Create the virtual environment

  python -m venv env or python3 -m venv env

• Activate the virtual environment

  cd .\env\Scripts\activate

• You should see a (env) in yout terminal

2.) Install the Packages

• Navigate back to the root directory

      cd ..
      cd ..

• Install the pip packages from the requirements.txt

  pip install -r requirements.txt
  
  

Steps Taken

1. Created and Ran init.py which:

   • Creates the CSV: NHL_teams_historical_stats_20132014_to_20222023.csv - used for creating and training the model.
   • Creates the CSV: NHL_teams_historical_stats_20232024.csv - used for testing the model and prediction.
   • Creates the CSV: NHL_teams_schedule_20232024_2024-02-26.csv - used for prediction.

2. Created and ran model.py, which creates three Classification models:

   • Random Forest (RFC)
   • Support Vector Machines (SVM/SVC)
   • Multilayer Perceptron Classifier (MLP)

   For each model:

   • Imported the NHL_teams_historical_stats_20132014_to_20222023.csv
   • Dropped columns that are not needed and those that cause data leakage.
   • Updated PHX to ARI as they have changed their team name (special case).
   • Created a new feature column for days since last game for each team.
   • Created a new feature column for the current win streak for each team.
   • Created a new feature column for the games played for each team.
   • Saved the cleaned data to a csv: NHL_teams_historical_stats_20132014_to_20222023.csv_cleaned.csv
   • One-hot encoded 'homeRoad', 'opponentTeamAbbrev', 'teamFullName'
   • Fill in missing or N/A values with 0.
   • Split the data into train and test sets.
   • Created the 3 classifier models (RFC, SVM, MLP).
   • Ran the test set, which prints model metrics to the console.
   • Saved the prediction model and normalization values for each model to a .joblib file in the respective output folder (allows it to be used in the future without retraining).

3. Created and ran test.py which:

   • Tests the model against:
     • current season completed games (up to: feb 26, 2024).
     • remaining games for the current season (feb 27 - Apr 18).

4. Results are output to csv/test

   • csv/test/current_season_predictions.csv
     • current season completed game results.
     • this is unseen data for each model.
     • accuracy is very similiar to the training accuracy.
   • csv/test/2023_2024_02-26_predictions.csv
     • predicts game results for each remaining game left in the season
     • feature values are calcualted using averages where necessary