1. NHL has 31 teams (Vegas Golden Knights joined in 2017)
2. Scoring system

• 0 points for a loss
• 1 point for a loss in overtime or a shootout
• 2 points for a win

1. My datasets are from 2015-2016, 30 teams are included
2. Extract the components from the scaled train values
   
3. Determine the number to retain
4. Rotate the retained components - orthogonal
5. Interpret the rotated solution: use principal() function
6. Create the factor scores - rotated components (RC1, RC2, ..)
7. Use scores as input variables for linear regression analysis
   
8. Evaluate the performance on the test data - predict

• Use of library(psych) as the Principal Components Analysis (PCA) package to so simplify datasets in supervised learning
• Orthogonal rotation (called "varimax") and interpretation to maximize the loadings of the variables on a specific component:

pca_rotate <- principal(train_scale, nfactors = 5, rotate = "varimax")

• Create factor scores for each team from the components

pca_scores <- data.frame(pca_rotate$scores)

• Use of linear regression to receive back the predictive scoring in machine learning lm() and predict()

• ppg: points per game

Links & References

• Hockey Reference
• NHL Statistics
• Open Source Sports - Hockey
• Principal component analysis
• Final Project on Github