多产球员的挖掘：超越常规措施

发布时间：2018-02-12 20:26

  本文关键词： 板球 运动数据挖掘 预测 排序 机器学习　出处：《北京邮电大学》2017年博士论文　论文类型：学位论文

【摘要】：体育领域的趋势分析方法已经从传统的简单的数据统计发展到当今的基于数据挖掘的深度分析。在一些主流的体育运动中,对板球态势的分析仍然非常落后,这引起学者们的广泛关注。为了填补球员预测和等级划分等方面的空白,同时消除当前传统统计方法的局限性,基于数据挖掘的态势分析作为一种非常有效的方法应运而生。数据挖掘的效率和准确性等方面的优势,日益凸显。到目前为止,板球相关的解决方案中大都用到了正交化统计和优化理论,而没有使用数据挖掘。板球组织正在征求可以融合多种方法的有价值的度量标准和机制,用以提高决策的有效性。此外,这样的分析对于权威中心是非常有益的,例如教练和管理者在提高专业技能时,能够获得球员和球队的最佳表现。论文旨在提供有效的解决方案,解决板球领域的之前没有考虑的问题以及当前解决方法仍然受限的问题。更准确地说,通过整合先进的数据挖掘工具(包括监督机器学习和随机游走算法),提供板球运动的有效解决方案。本文的创新点如下:·基于机器学习的方法,提出了预测新星的综合性能评价函数,能够鲁棒准确地预测新星。我们首次整合协作球员、团队以及敌手的概念,并通过实验的方法针对击球手筛选出Co-batsmen Runs和Co-batsmen Average 等 9 方面的潜在特质 ,对投球手筛选出 Team Average和Team Strike Rate等11方面的特质,然后对这些特质采用同类聚合归一的方法进行公式化。为了研究分类,采用了生成性和判别性的机器学习算法。交叉验证表明算法可以高精度的预测新星,并且该算法不但具有鲁棒性,而且统计效果显著。·基于树型监督机器学习方法,提出了预测明星球员的贝叶斯模型,取得了好的预测结果。我们首次利用球员的年智发展状况提取若干击球和投球特性,并找到主导地位的个体特征。在合并测试集上筛选出六种合适的监督机器学习分类算法用于明星球员的二元分类,并通过整合使用贝叶斯定理、函数以及基于树的监督机器学习算法给出融合的算法。最终,利用该融合算法对贝叶斯机制、树形机制、函数机制下的两类模型进行评估,实验结果证明该算法具有出色的鲁棒性和高效性。·综合投球、击球和团队优先级等信息,提出了团队得分能力的预测算法,实验结果表明该算法具有较高的预测精度。在球队层面,投球手和打球手的能力与输赢密切相关,并且依据此能力来选拔出最强的团队。针对以前的工作使用特性较少的情况,我们提出新的预测评估算法,卷入更多的属性,使得该算法具有以下特点:一、在球队得分能力排名方面,我们的算法更高效;二、突破以前算法的限制,高效预测击球、投球得分能力;三、通过获胜的球队得分能力优先次序与世界杯各自冠军的冠军进行了交叉检查,发现各个比赛跨度期间最有得分能力的队伍未必赢得世界杯。
[Abstract]:The trend analysis method in the field of sports has changed from the traditional simple statistics to depth analysis based on data mining today. In some mainstream sports, analysis of the situation of cricket is still very backward, which attracted the attention of scholars. In order to fill the blank player prediction and classification, at the same time to eliminate the current limitations of traditional statistical methods, data mining analysis of the situation as a kind of very effective method based on data mining came into being. The efficiency and accuracy of advantages, has become increasingly prominent. To date, the most relevant cricket solutions used orthogonal statistics and optimization theory, without the use of data mining the organization is seeking. Cricket can integrate a variety of methods of value measure and mechanism is effective to improve the decision. In addition, the analysis of the In the centre of authority is very useful, for example, coaches and managers in improving professional skills, can obtain the best players and teams. This paper aims to provide effective solutions, the cricket field did not consider before solving problems and the solution method is still limited. More precisely, through the integration of advanced data mining tools (including supervised machine learning and random walk algorithm), to provide effective solutions to cricket. The innovation of this paper is as follows: Based on machine learning method, the performance evaluation of the star prediction function is proposed, which can accurately predict the robust star. We first integrated collaborative team player and opponent concept. And through the experiments for the batter selected Co-batsmen Runs and Co-batsmen Average latent trait in 9 aspects, Team Average of the bowler screening Team Strike and Rate 11 characteristics, and then the formula of these traits by using the method of similar polymerization normalization. In order to study the classification, using the generative and discriminative machine learning algorithm. Cross validation shows that the algorithm can predict the star high precision, and the algorithm is not only robust, but also statistical significant effect based on the tree. The supervised learning method, proposed the Bias prediction model of star players, achieved good prediction results. We first use of years of wisdom to develop players picking batting and pitching characteristics, individual characteristics and find the leading position in the combined test set selected six kinds of machine learning supervision right classification algorithm for classification of two yuan star players, and through the integration of the use of Bias's theorem, function and algorithm is given based on the fusion algorithm of supervised machine learning tree finally. And using the fusion algorithm of Bayesian tree mechanism, mechanism, two kinds of model function under the mechanism of evaluation, experimental results show that the algorithm has good robustness and efficiency. - pitching, hitting and team priority information, put forward the calculation method of pre team scoring ability, the experimental results show that the prediction accuracy of the algorithm high. In the team level, and winning pitcher and hitter is closely related to, and on the basis of the ability to select the strongest team. According to the characteristics of the previous work using less, we propose a new algorithm to forecast and evaluate, involving more attributes, the algorithm has the following characteristics: first, in terms of ranking in the team scoring ability, our algorithm is more efficient; two, to break the previous algorithm, efficient prediction of batting, pitching score ability; three, by winning the first scoring team order A cross examination with the champions of the world cup has been conducted to find that the team who has the most scoring ability during the span of the game does not necessarily win the world cup.

【学位授予单位】：北京邮电大学
【学位级别】：博士
【学位授予年份】：2017
【分类号】：TP311.13;G80-3
，

本文编号：1506487