職業運動競賽陣容對決勝負預測-以國際籃球協會為例

Predicting the Outcome of the Professional Sports Based on NBA

林哲琛(Che-Chen Lin)

指導教授 ： 孫春在

國立交通大學/資訊學院/資訊科學與工程研究所/碩士(2018年)

摘要

各種不同的陣容組合產生的勝負情形是難以解開的答案，由於陣容的組成相當複雜，難以從直觀的角度對結果做出判斷，勝負的結果對於賭注會有相當重大的利益關係，這也造成許多人的關注與辯論，而在大眾的討論下對於整個網路社交又會造成非常重大的影響，這也成為一個難以去預測的問題，多數人常會以自身的經驗去做預測判斷，不一定會以客觀的角度去做評估，各大媒體平台都可以發現相關的話題，造成國際籃球協會的關注程度獲得提升，對於體育聯盟的發展趨勢也產生相當巨大的衝擊。
本研究想要針對陣容之間產生的勝負情形進行預測，將利用NBA真實賽場的資料，並且以NBA 2K這款遊戲的球員能力值當作球員的資訊，藉由許多針對籃球的研究選擇訓練方法以及對資料做前處理，接著輸入類神經網路以監督式學習的方式進行訓練，本研究的實驗結果對於得分差越大的資料可以得到更為準確的預測，而實驗顯示對於籃球場上無法以數據呈現的球員能力也是相當重要。
本研究的結果雖然在資料的數量上相當有限，然而還是可以得到一定的成效，並且得知球員無法數據化的能力是相當重要的，這對於很多團體競賽方面的陣容對決也能夠以本研究的方式達到成效，根據不同的陣容對決情形可以輸入相關領域的重要屬性特徵，可以用同樣的方式進行勝負的預測。

關鍵字

人工智慧 ； 深度學習 ； 類神經網路 ； 籃球 ； 預測

並列摘要

There is a very huge discussion topic in the outcome of the matchup due to the lineup duels. Because the composition of the lineup is quite complex, it is difficult to predict the results from an intuitive perspective. The outcome has a significant impact on the gambling. This also caused many people's attention and debate. And under the discussion of the public, it will have a very significant impact on the entire social network. This also becomes a difficult problem to predict. Most people often use their own experience to make predictions and do not make predictions in an objective way. We can find related topics on many media platforms and caused an increase in the attention of the International Basketball Association. There is also a huge impact on the development of sports leagues.
This study is aimed at predicting the outcome of the lineup duel. It will use the real NBA games data and use the NBA 2K players’ ability as NBA players’ information. Select training methods and handle data with many researches for basketball, and then input data into neural network to supervise learning. The results of this study can provide more accurate predictions for data with larger score differences, and result show that players’ abilities that cannot be recorded on the box score are also very important.
Although the results of this study are quite limited in the number of data, they can still achieve certain results. The abilities that cannot be recorded on the box score are also important. This research can also achieve results for the different team competition duels. According to the different lineup duels, you can input important attribute characteristics of the relevant field, and use the same way to predict the outcome.

並列關鍵字

Artificial Intelligence ； Deep Learning ； Neural Network ； Basketball ； Prediction

參考文獻

[1] J. B. Yang, and C. H. Lu, ” Predicting NBA championship by learning from history data,” Proceedings of artificial intelligence and machine learning for engineering design, 2012.

[2] D. Miljković, L. Gajić, A. Kovačević, and Z. Konjović, ” The use of data mining for basketball matches outcomes prediction,” Intelligent Systems and Informatics (SISY), 2010 8th International Symposium on. IEEE, pp. 309-312, Nov, 2010.

[3] B. Loeffelholz, E. Bednar, and K. W. Bauer, “Predicting NBA games using neural Networks,” Journal of Quantitative Analysis in Sports, vol. 5, pp.1-17, Jan, 2009.

[4] D. Delen, D. Cogdell, and N. Kasap, ” A comparative analysis of data mining methods in predicting NCAA bowl outcomes,” International Journal of Forecasting, vol. 28, pp. 543-552, Apr, 2012.

[5] M. C. Purucker, “Neural network quarterbacking,“ IEEE Potentials, vol. 15, pp. 9-15, Aug, 1996.