Machine Learning Outperforms Logistic Regression Analysis to Predict Next-Season NHL Player Injury: An Analysis of 2322 Players From 2007 to 2017
Luu BC, Wright AL, Haeberle HS, Karnuta JM, Schickendantz MS, Makhni EC, Nwachukwu BU, Williams R, Ramkumar P. Machine Learning Outperforms Logistic Regression Analysis to Predict Next-Season NHL Player Injury: An Analysis of 2322 Players From 2007 to 2017. Orthopaedic Journal of Sports Medicine 2020; 8(9).
Orthopaedic Journal of Sports Medicine
© The Author(s) 2020. Background: The opportunity to quantitatively predict next-season injury risk in the National Hockey League (NHL) has become a reality with the advent of advanced computational processors and machine learning (ML) architecture. Unlike static regression analyses that provide a momentary prediction, ML algorithms are dynamic in that they are readily capable of imbibing historical data to build a framework that improves with additive data. Purpose: To (1) characterize the epidemiology of publicly reported NHL injuries from 2007 to 2017, (2) determine the validity of a machine learning model in predicting next-season injury risk for both goalies and position players, and (3) compare the performance of modern ML algorithms versus logistic regression (LR) analyses. Study Design: Descriptive epidemiology study. Methods: Professional NHL player data were compiled for the years 2007 to 2017 from 2 publicly reported databases in the absence of an official NHL-approved database. Attributes acquired from each NHL player from each professional year included age, 85 performance metrics, and injury history. A total of 5 ML algorithms were created for both position player and goalie data: random forest, K Nearest Neighbors, Naïve Bayes, XGBoost, and Top 3 Ensemble. LR was also performed for both position player and goalie data. Area under the receiver operating characteristic curve (AUC) primarily determined validation. Results: Player data were generated from 2109 position players and 213 goalies. For models predicting next-season injury risk for position players, XGBoost performed the best with an AUC of 0.948, compared with an AUC of 0.937 for LR (P <.0001). For models predicting next-season injury risk for goalies, XGBoost had the highest AUC with 0.956, compared with an AUC of 0.947 for LR (P <.0001). Conclusion: Advanced ML models such as XGBoost outperformed LR and demonstrated good to excellent capability of predicting whether a publicly reportable injury is likely to occur the next season.