Application of Memory-Based Collaborative Filtering to Predict Fantasy Points of NFL Quarterbacks

Abstract

Subjective expert projections have been traditionally used to predict points in fantasy football, while machine prediction applications are limited. Memory-based collaborative filtering has been widely used in recommender system domain to predict ratings and recommend items. In this study, user-based and item-based collaborative filtering were explored and implemented to predict the weekly statistics and fantasy points of NFL quarterbacks. The predictions from three seasons were compared against expert projections. On both weekly statistics and total fantasy points, the implementations could not make significantly better predictions than experts.However, the prediction from the implementation improved the accuracy of other regression models when used as additional feature.