MAYFIELD: Machine Learning Algorithm for Yearly Forecasting Indicators and Estimation of Long-Run Player Development

Download the
Full Paper Here
Authors

Alexander Williams, The Ohio State University
Sethward Brugler, General Motors
Benjamin Clarke, The Ohio State University

Abstract

Accurate statistical prediction of American footballplayer development and performance is an important issue in the sportsindustry. We propose and implement a novel, fast, approximate k-nearestneighbor regression model utilizing locality-sensitive hashing in highlydimensional spaces for prediction of yearly National Football League playerstatistics. MAYFIELD accepts quantitative and qualitative input data, and canbe calibrated according to a variety of parameters. Concurrently, we proposeseveral new computational metrics for empirical player comparison andevaluation in American football, including a weighted inverse-distancesimilarity score, stadium and league factors, and NCAA-NFL statisticaltranslations. We utilize a training set of comprehensive NFL statistics from1970-2019, across all player positions and conduct validation on the model withthe subset of 2010-19 NFL statistics. Preliminary results indicate the model tosignificantly improve on current, publicly available predictive methods. Futuretraining with advanced statistical datasets and integration with scouting-basedmethods could improve MAYFIELD's accuracy even further.