Ohio State nav bar

Student Research Winner: Williams, Clarke and Brugler on "MAYFIELD: Machine Learning Algorithm for Yearly Forecasting Indicators and Estimation of Long-run Player Development"

Abstract

Accurate statistical prediction of American football player development and performance is an important problem for sports managers, executives, media, and fans. We propose and implement a novel, fast, approximate k-nearest neighbor regression machine learning model utilizing locality-sensitive hashing in highly dimensional Euclidean spaces for prediction of year-to-year National Football League player statistics. The algorithm accepts quantitative and qualitative input data, and can be calibrated according to a variety of parameters. Concurrently, we propose several new computational metrics for empirical player comparison and evaluation in American football, including a weighted inverse-distance similarity score, stadium and league factors, and NCAA-NFL statistical translations. We utilize a training set of elementary NFL quarterback statistics from 1970-2017, although the algorithm can be expanded to other positions and statistics. Furthermore, we conduct cross-validation on the model with a set of 2018 NFL quarterback statistics. Preliminary results indicate the model to be an improvement over current, publicly available predictive methods. Although computationally intensive, future work and training with expanded datasets could improve our algorithm even further.

Sports and Society Statement

Managers of sports teams, as well as fans, sports media, and others, need better tools for accurate and stable evaluation of player performance. This project proposes a novel method for the prediction of NFL player statistics, and advances several other new descriptive metrics of historical player performance. Potential applications for MAYFIELD could include use by players to negotiate for higher salaries if they have favorable stats projections, and use by collegiate athletes to assist their decision of when to declare for the draft (e.g., a junior with a remaining year of eligibility could consult his MAYFIELD forecast to see its estimates of his preparedness to play professionally). Due to the public availability of MAYFIELD's methodology and results, our work may also inspire additional advancements in sports analytics by others which build off the techniques we use for MAYFIELD.