以迴歸樹預測美國職棒大聯盟各球隊的年度勝率及晉級季後賽之名單

Predicting Yearly Winning Percentage and Playoff List of MLB Teams by Regression Trees

羅莉雯(Li-Wen LO)

指導教授 ： 周清江

淡江大學/商管學院/大數據分析與商業智慧碩士學位學程/碩士(2019年)

https://doi.org/10.6846/TKU.2019.00190

摘要

棒球是世界上最受歡迎的運動之一。根據谷歌趨勢分析，過去五年中，觀看美國棒球大聯盟（MLB）比賽的人越來越多。許多學者和球迷都對預測比賽結果有很大的興趣，他們使用球隊的各項表現來預測比賽結果。過去的研究預測準確率約為50%左右。然而，有一些球隊在確定進入季後賽後讓主力球員休息，以替補球員為主進行比賽。因此，本研究僅使用2016到2018各年度前半季各球隊表現的平均來預測年度勝率。本研究的方法分別為分類和迴歸樹（CART）、最大概似迴歸樹（MLRT）、分類相關迴歸樹（CCRT）和最大概似相關迴歸樹（MCRT）。本研究使用平均絕對百分誤差（MAPE）作為評量模型依據，所有模型的預測MAPE皆在10％到15％之間。最後，我們將預測的勝率應用於預測大聯盟季後賽名單及預測各分區第一名。由於MAPE無法將球隊排名加入計算，本研究使用平均精度均值（MRR）於預測各分區第一名。季後賽名單及預測各分區第一名的最高預測準確率皆為88％。根據本研究結果顯示MLRT的性能優於CART。在季後賽名單預測中，本研究結果也優於先前之研究。實驗顯示，本研究使用的方法可用於預測球隊的年度勝率、MLB的季後賽名單及各分區第一名預測。

關鍵字

美國職棒大聯盟 ； 迴歸樹 ； 迴歸分析 ； 最大概似迴歸樹 ； 分類迴歸樹 ； 赤池信息量準則 ； 平均絕對百分比誤差 ； 平均精度均值 ； 預測勝率 ； 預測季後賽名單

並列摘要

Baseball is one of the most popular sports in the world. According to Google Trends analysis, over the past five years, more and more people are watching games of Major League Baseball (MLB) of the USA. Many scholars and fans are interested in predicting these games’ outcome based on the teams’ yearly performances. The prediction accuracies of previous studies were around 50 percent. However, some of teams do not play seriously after they are sure to enter the playoff. Thus, instead of prediction based on a team’s performance of the whole season, we used a team’s performance of the first half season to predict its yearly winning percentage. We apply Classification and Regression Trees (CART), Maximum Likelihood Regression Trees (MLRT), Classification and Correlation Coefficient Regression Trees (CCRT) and Maximum Likelihood and Correlation Coefficient Regression Trees (MCRT) separately in the study. Our prediction error of the winning percentage in terms of Mean Absolute Percent Error (MAPE) is between 10 to 15 percent. Then we apply the predicted results to obtain the playoff list. And we use Mean Reciprocal Rank (MRR) in predicting the best team in each region. Our highest prediction accuracy of the playoff list and the best team in each region are 88 percent. Our results show that performance of MLRT is better than that of CART. In predicting playoff list, our study's result is better than previous research. Experiments demonstrate that our used methodology could be used to predict a team’s yearly winning percentage, playoff list and the best team in each region of MLB.

並列關鍵字

MLB ； Regression tree ； Regression analysis ； MLRT ； CART ； AIC ； MAPE ； Mean Reciprocal Rank ； Winning percentage prediction ； Playoff list prediction

參考文獻

[1] Akaike, H., A new look at the statistical model identification, IEEE Transactions on Automatic Control, 1975, Vol. 19, No. 6, p.716-723.

[2] Benesty, J., Chen, J., Huang, Y., Cohen, I., Noise Reduction in Speech Processing, 2009, p.37-40.

[3] Brieman, L., Friedman, J., Olshen, R., and Stone, C., Classification and Regression Trees, Belmont, CA: Wadsworth International Group, 1984.

[4] Chaudhuri, P., Loh, W., Nonparametric estimation of conditional quantiles using quantile regression trees, Bernoulli, 2002, Vol. 8, No. 5, p. 561–576.

[5] Chen, C., Developing winner prediction models of professional baseball using data mining techniques, Master thesis of National Taiwan Sport University (in Chinese), 2011.