National Hockey League prospect evaluation: utilizing machine learning to predict amateur prospect career success

Abstract

Evaluation of prospect potential at a professional level is crucial to building a championship roster in any sport. In the National Hockey League (NHL), prospect evaluation is just as critical, but even more difficult than in other sports, due to the plethora of amateur leagues globally and the youth of most prospects. This report summarizes the process of collecting North American amateur skater data between the 2007 and 2024 hockey seasons, alongside player draft position, and corresponding career expected goals in the NHL, all obtained to analyze predictive accuracy of several machine learning algorithms in predicting career expected goals. Player physical attributes, such as height and weight, were incorporated with amateur statistics from their final season before being drafted as predictor variables. Overall, bagged trees and random forest modeling had the smallest error in expected goal predictions on the test dataset. Such a model could be further developed and used as an evaluation tool for predicting prospects' NHL career success.