Hockey analytics has created a large array of statistics to measure player performance. How to properly weigh each of these to create a comprehensive overview of each player is not as widely discussed. Some, including myself, have tried to smartly roll them up into a single metric. However, possibly more appropriately, they can also be used to perform a clustering analysis, identifying player types.

To do this, it helps to first reduce the number of features we have for each player. This is best done with Principal Components Analysis (PCA), which boils down all metrics into a few by creating a ‘Principal Component’ based on finding the maximum variability in the dataset. Then finding the second dimension of maximum variability, and so on.