AN INVESTIGATION INTO UTILIZING PLAYER PERFORMANCE TRAITS AS A FRAMEWORK FOR A MULTIVARIATE ALGORITHM INTENDED FOR PREDICTING FOOTBALL GAME OUTCOMES

Abstract

The financial sector has seen an increasingly large amount of money being poured into machine learning algorithms and big data to predict seemingly random events such as loan defaults which fundamentally relies on human traits and behaviors. Sports competition outcomes have often been considered completely random due to their innate reliance on players who have unpredictable abilities to perform from game to game. Modern technology and a well-financed simulation video game industry have allowed for a large-scale database of NFL players which numerically quantifies a multitude of player traits to simulate their on-field success. This study will seek to validate the increasingly plausible idea that seemingly random football game outcomes can be correctly predicted at a rate better than 50% by quantifying the underlying players’ talents of each opposing team and using this data to create multiple (albeit generalized) composite scores for each team. These composite scores will be weighted and compared between opposing teams to predict a game outcome.