Mixed Clustering Methods to Forecast Baseball Trends

Abstract

Sport betting has become one of the most profitable business around the world. This business generates millions of dollars every year. One of the most influenced games is Baseball. Baseball has suffered an important change after the introduction of statistical methods to tune up the team strategy. This effect, called Moneyball, started in 2002 when the team Oaklans Atletics began to choose players according to their statistics. After this successful approach, several teams decided to continue with this strategy, generating strong statistical teams. The statistical information about players and matches have acquired highly importance, creating different datasets, such as Retrosheet which collects detailed information about players, teams and matches since 1956 until today. This work pretends to generate a forecasting model for Baseball focused on the result prediction of new matches using statistical previous information. We combine time-series and clustering algorithms to generate a model which learns about the teams and matches evolution and tries to predict the final results. Even whether this model is not complete accurated, it becomes a good starting point for future models.