- 學位論文
利用深度學習進行運動賽事結果預測
Applying Deep Learning to Forecast the Results of Sports Events
摘要
從國際世足賽及世界棒球經典賽的風潮席捲而來之時，從報章媒體或是網路等其他媒體中的討論程度；就能夠發現這些競技類型的球賽運動，對於大多數人們來說有著非常重要的地位，同時也是討論性很高的話題；甚至有許多的人只是一日球迷，而這些人們都會有著相同的期望；就都是希望自己所支持的隊伍能夠勝出。 本研究將以同為球類競技賽事的國際籃球協會（National Basketball Association）進行分析及探討；使用NBA過去的比賽紀錄及數據來進行比賽結果的預測。此研究將結合深度學習（Deep learning）方法，使用卷積神經網路（Convolutional Neural Networks）及過去3年總共4235場的比賽數據做為分析資料給予模型進行訓練及預測；之後依照不同架構的模型所訓練得出的結果進行探討及分析。 在過去的研究中，卷積神經網路較多使用於圖像辨識或者是物體辨識等研究，而本研究提出一個編碼方式並透過結合深度學習的方式進行比賽結果預測，其模型的預測精準度大約為91%，而誤差值約莫為0.2左右，其結果證實本研究所提出的編碼方式的可能性。
並列摘要
The importance of competitive sports events such as the World Cup and the World Baseball Classic for a majority of people can be easily found through the heated discussions in newspapers and other types of media such as the Internet while the fad hits. They are also highly discussed topics. Many people are even one-day fans with the same expectations; that is, they want their team to win. In this study, records and data from the many contests that the National Basketball Association (NBA), which also deals with competitive sports, has held will be analyzed and discussed in order to forecast results of games. The deep learning approach will be adopted and convolutional neural networks and data from 4235 games over the past 3 years will be used for analysis and to facilitate training on and forecasts done applying the model. Finally, forecast results obtained applying the model will be discussed. In prior studies, convolutional neural networks were more frequently applied to identifying images or objects. This study proposes a encoding method combine deep learning to forecast of event results. The accuracy of the CNN model is about 91%, and the loss is about 0.2. The results confirm the possibility of the coding method proposed in this study.