| Abstract |
The purpose of this study is to find the critical factor affecting outcomes of baseball games. The routine games from 2008 to 2011 between LAD and SF of MLB are used as examples. There were 18 games per year, a total of 72 games, of which 54 are adopted as training data of neural network model for predicting the last 18 games in 2011 and the games held on May 8th, 9th and 10th of 2012. Thus a total of 21 games are the primary test data. This study analyzes and integrates the data recorded after each routine game between LAD and SF, established the critical factors affecting game outcomes, and uses neural network technology to predict the winning team. In aspect of offense, critical factors include HR, RBI, AVG and OBP; in aspect of pitcher, factors include IP, HR, AVG and WHIP. The neural network model adopts the back-error propagation neural network model for its stability, and easiness to converge and learn. This model is compared with other models which applied statistical methods, and the results show that the neural network model is viable for prediction of non-linear models, and can be used to understand the extent of impact of these critical factors on the games. |
