프로야구 포스트시즌 진출 예측을 위한 통계적 모형 비교

Abstract

The purpose of this study is to compare the outcomes of three statistical methods being used frequently in predicting the post-season entry of professional Korean baseball. The data were collected from official materials issued by KBO and related organizations, and consisted of the records of each individual skills of professional baseball played in Korea from 1991 to 2009. Based on the data obtained in regular season three statistical methods (discriminant function(DFA), logistic(LOS), and neural network analyses(NNA) were applied to predict the outcome of post-season entry, and the results of 3 statistical models and the outputs of SPSS Windows 16.0 ware compared in terms of selected predictors, hit rates, and statistical goodness of fit indices. The following results were obtained from the analyses performed in this study. First, the mean value of each skill performance for those teams who made post-season entry was considerably higher than those who did not. This was particularly so in those individual skills related to pitching and hitting skills. Second, the performance of DFA and LOS was pretty much the same in pitching (80.3% and 80.2%, respectively), in hitting/scoring (88.2%, 88.8%), and overall ability (89.5%, 89.5%) in term of the percent of hit rate. However, the percent of hit rate for NNA was a bit hitter than those of two other methods in each of the skill areas, but the differences were minimal. However, this study suggested that the selection of independent variables in the prediction of the outcome of sport performance should not be solely relied upon the statistical criteria, and thus theory-based hierarchical approach should be considered in order to obtain an external validity of the variables selected for the predictors in regression related techniques.