Summary: | 博士 === 輔仁大學 === 商學研究所博士班 === 104 === The purpose of this study was to develop the Fraud Detection Model for Professional Baseball Games (FDMPBG). The researcher used data mining techniques to build the FDMPBG by reviewing 6,628 games from the season records of 1990 to 2015 in the Chinese Professional Baseball League (CPBL) and the final verdicts of the match-fixing games among CPBL from the high courts in Taiwan. Due to the serious imbalance between the match-fixing games and the nomal games of CPBL, the under-sampling based on clustering (SBC) was applied to reduce this situation. The FDMPBG was built by using discriminant analysis (DA), logistic regression analysis (LR), artificial neural networks (ANNs), multivariate adaptiveregression splines (MARS) and support vector machine (SVM). This study also investigated the important variables for the match-fixing games of CPBL and compared the strengths and weaknesses of different models. Based on the evaluation index of accuracy, sensitivity, specificity, precision, recall and F-measure, the findings of this research verified the SVM was better than other data mining techniques.
|