職棒賽事勝負之預測－以中華職棒二十八年賽季為例

Preditct the Results of Professional Baseball Games- Take the 28th season of Chinese Professional Baseball League for Example

林清榆(Cing-Yu Lin) ； 蔡典龍(Tien-Lung Tsai)

《休閒運動保健學報》 15期 (2018/12) Pp. 1-16

https://doi.org/10.6204/JRSHP.201812_(15).0001

摘要

本研究目的為預測職棒比賽結果，研究對象為中華職棒大聯盟2017年賽季（中華職棒二十八年）的四支球隊，將採用三種解釋變數處理方式，為打擊成績與投手成績、打擊成績比與投手成績比、打擊成績差與投手成績差，在分別對4個時間點（賽季進行20%、賽季進行40%、賽季進行70%、賽季進行90%）建立邏輯斯迴歸模型，再對其後10場比賽以6種預測方式進行預測驗證。選擇解釋變數的方式，採用Wald統計量逐步選取。另外將賽季70%、90%分別對下半季20%、40%進行比較，也針對全賽季前一場與前面全賽季的三種解釋變數處理方式對照，判別其中差異。研究結果三種解釋變數處理方式對6種預測方式的最高平均預測成功率分別是70%、72.5%、67.5%，其中全賽季70%、90%略優於下半季20%、40%，而前一場的處理方式預測成功率也未必比這三種解釋變數處理方式好；預測方式則是以兩隊上次交手數據做為解釋變數時，預測成功率在三個解釋變數處理方式中都是最好的；選入變數方面，包含下半季及前一場共選入了22種變數，其中最高的三個為每打數平均打點、防禦率、三振四壞比。

關鍵字

邏輯斯迴歸 ； 職業棒球 ； 預測

並列摘要

Forecasts the Professional baseball competition result the object for the Chinese Professional Baseball League in 2017 the season (CPBL 28 years) four teams, we will use three explanation variable processing mode, that is batting score and pitching score, the division of batting score and pitching score, and the subtraction of batting score and pitching score, and establish the Logistic regression model for 4 times (20% for the season, 40% for the season, 70% for the season, 90% for the season), then predict and verify the last 10 games by 6 methods of prediction. Use Wald test to choose the explanation variable, and compare the 70% and 90% for the season with 20% and 40% for the bottom season, and also control the 3 explanation variable for the previous game to all season and the previous season. The research result is that the highest average predicted success rate is 70%,72.5% and 67.5% for 3 explanation variable to predict the 6 methods of prediction, in which 70% and 90% for all season is better than 20% and 40% for bottom season, but the preceding processing mode forecast success ratio not necessarily is also better than these three explanation variable processing mode; Forecast the way is previous time fights by two rows when the data does for the explanation variable, the forecast success ratio in three explanation variable processing mode all is best; selects into the variable aspect, contained next half Ji Jiqian altogether to select into 22 kind of variables, in which highest three has RBI,ERA and SO per bat.

並列關鍵字

Logistic regression ； professional baseball ； predict