- 學位論文
用資料探勘方法預測對抗型團隊比賽的勝率－以NBA例行賽為例
Using Data Mining Method to Predicting Winning Percentage for Dual Meet Team Sport – Using NBA Regular Season as a Case Study
摘要
本研究目的在於找出影響對抗型球隊比賽勝負的關鍵因素，並藉由建立適合NBA數據的模型，進而以此構建對抗型球隊比賽的數據模型，預測未來競賽的勝負情況。研究中使用了1996-1997年賽季到2015-2016年賽季資料，並產生預測2016-2017年賽季的得分與勝負，再與實際賽事結果做比較，以此判斷模型的適用性。 本研究依據建模之後的結果，得到兩分得分與罰球得分是影響球隊勝負的關鍵因素。使用預測所得到的分數來判斷模型的準確率，其結果顯示這四個模型 (rpart、cubist、randomforest、svm) 的錯誤率均小於12%，預測準確率均在60%上下波動。更進一步來說，cubist和randomforest兩個模型在2013-2016這3個賽季與2012-2016這4個賽季中的錯誤率皆小於10%，表示這兩個模型具有高準確的預測能力。在預測勝負方面，研究結果顯示第四次的測試準確率都較其他測試結果佳，準確率均介於56到60% 之間。從此研究中發現，在做資料分析或預測時要使用較多的模型相互比較，才能做較客觀的判斷。另外，根據本研究數據，數值型的預測能力強過分類型的預測能力。
並列摘要
The purpose of this study is to find out the key factors that affect the outcome of the confrontational team competition, and to build the model for the NBA data, and then build the data model of the opposing team competition to predict the outcome of the future competition. The use of the 1996-1997 season to 2015-2016 season data, and produce the forecast 2016 - 2017 season, the score and the outcome, and then compared with the actual results to determine the applicability of the model. Based on the results after modeling, the two points score and free throw score is the key factor affecting the outcome of the team. The results show that the error rate of these four models (rpart, cubist, randomforest, svm) is less than 12%, and the prediction accuracy fluctuates at 60%. Further, the Cubist and randomforest models have less than 10% error rates in the three seasons of 2013-2016 and 2012-2016, indicating that the two models have high accuracy. In the prediction of the outcome, the results show that the fourth test accuracy than other test results are good, the accuracy rate is between 56 to 60%. From this study found that in the analysis or prediction of data to use more models to compare each other in order to make more objective judgments. In addition, according to the data of this study, the numerical predictive ability is stronger than the predictive ability.