Joint work with John Yannotty.

GitHub: View project on GitHub

Fantasy football has become a prominent feature in modern day society, as millions of people globally invest themselves both emotionally and financially. Player scoring projections and rankings play integral roles, as each provides fantasy team owners with crucial information to make week by week decisions. This study focuses on evaluating both of these measurements through statistical learning methods. By this, classification models are built to predict a player’s relative scoring performance. Furthermore, a soft ensemble model is constructed in conjunction with multiple individual models. Finally, a hierarchical ranking of players is computed through k-means clustering.

To date, most models are driven by relevant player and team statistics. The proposed model in this study couples these general practices with twitter sentiment by which tweets referring to specific players are analyzed to quantify the manner in which the player is being viewed by both fans and analysts.

Models were constructed for quarterbacks and running backs and resulted in roughly 70% classification accuracy. Final player rankings were compared to current ESPN player rankings and the top tier for QB’s and RB’s displayed 62.5% and 100% agreement respectively.

For additional details, please contact author.