Baseball Prediction Using Ensemble Learning by Arlo Lyle

As the salaries of baseball players continue to skyrocket and with the ever-increasing popularity of fantasy baseball, the desire for more accurate predictions of players’ future performances is building both for baseball executives and baseball fans. While most existing work in performance prediction uses purely statistical methods, this thesis showcases research in combining multiple machine learning techniques to improve on current prediction systems by increasing the accuracy of projections in several key offensive statistical categories. By using the statistics of players from the past thirty years, the goal of this research is to more accurately learn from this data how a player’s performance changes over time and apply this knowledge to predicting future performance. Results have shown that using machine learning techniques to predict a player’s performance is comparable to the accuracy seen by some of the best prediction systems currently available. Index words: Machine Learning, Ensemble Learning, Baseball Prediction, Model Trees, Artificial Neural Networks, Support Vector Machines, Bagging, Boosting, Stacking