基于最大熵原理的NBA赛事胜负预测与方法研究

发布时间：2018-03-17 02:15

本文选题：最大熵模型　切入点：Page　出处：《湘潭大学》2017年硕士论文　论文类型：学位论文

【摘要】：NBA作为全球最具影响力的篮球顶级赛事联盟,吸引着全世界篮球爱好者的关注,一直受大众追捧。球队之间的实力非常接近,以及大量随机性因素使得比赛结果的预测变得非常困难。考虑到比赛的比赛特征之间的相互影响,本文提出了一种基于最大熵原理的NBA赛事胜负预测方法,提高预测的正确率。首先,本文考虑到球队主力球员的变动对比赛结果的影响,对比赛双方已有的得分进行修改;将Page Rank算法的思想运用到修改后的历史比分数据中,构造球队之间的投票矩阵;然后利用幂法对投票矩阵进行求解,得到量化球队的真实相对实力值。其次,利用K-means聚类算法对所选取的代表比赛属性的特征数据进行离散化处理;并根据最大熵原理构造符合特征的NBAME模型。然后,在训练样本集上用GIS算法对NBAME模型进行最优化训练,得到模型中的参数的值,从而建立符合训练样本数据的NBAME模型。最后,将测试样本中比赛的特征数据代入到NBAME模型,计算对应比赛的主场球队获胜的概率,利用阈值来确定比赛的胜负。实验表明,当阈值取0.5时,模型能够预测所有测试集上的比赛,并且预测正确率最高能够达到75.6%;当阈值提高到0.7时,模型能够预测的比赛场次减少,但预测正确率却可以达到84.8%。本文所选取比赛的属性特征和依据最大熵原理构造出的NBAME模型能够有效的预测NBA比赛的胜负,相比现有的机器学习模型的正确率有所提升。
[Abstract]:NBA, the world's most influential league of top basketball events, has attracted the attention of basketball enthusiasts around the world and has long been sought after by the public. And a lot of random factors make it very difficult to predict the result of the competition. Considering the interaction between the characteristics of the competition, this paper proposes a method for predicting the results of NBA events based on the maximum entropy principle. First of all, considering the influence of the change of the main players on the result of the match, this paper modifies the existing scores of both sides of the game, and applies the idea of Page Rank algorithm to the revised historical score data. The voting matrix between teams is constructed, and then the voting matrix is solved by power method to get the real relative strength of the team. Secondly, the K-means clustering algorithm is used to discretize the selected feature data representing the game attributes. According to the principle of maximum entropy, the NBAME model is constructed. Then, the GIS algorithm is used to optimize the NBAME model on the training sample set, and the values of the parameters in the model are obtained. Finally, the NBAME model is established in accordance with the training sample data. The characteristic data of the match in the test sample is inserted into the NBAME model to calculate the probability of the home team winning the corresponding match, and the threshold value is used to determine the winning or losing of the match. The experiment shows that when the threshold is 0.5, The model can predict the matches on all test sets, and the prediction accuracy can reach 75.60.When the threshold is raised to 0.7, the number of matches predicted by the model decreases. However, the prediction accuracy can reach 84.8%. The NBAME model constructed according to the maximum entropy principle and the attribute feature of the selected competition can effectively predict the success or failure of the NBA match, which is improved compared with the existing machine learning model.
【学位授予单位】：湘潭大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：G841;TP181

【相似文献】