基于机器学习的NBA常规赛MVP投票分析与预测  

Analysis and Prediction of MVP Voting in NBA Regular Season Based on Machine Learning

作　　者：任昕[1] REN Xin(Dalian University of Science and Technology,Dalian 116052,China)

机构地区：[1]大连科技学院

出　　处：《竞争情报》2020年第1期9-23,共15页Competitive Intelligence

摘　　要：通过机器学习,利用1956—2018年NBA常规赛技术统计数据,对NBA常规赛MVP投票进行分析。运用线性相关性与变量重要性方法对球员的NBA常规赛技术统计变量进行分组与降维,发现“高级数据”中的变量与MVP得票率有显著的相关性,其中“胜利贡献值WS”的影响最大;同时发现由于1980年前后MVP投票制度发生重大变化,MVP得票率与变量之间的关系也发生明显变化,因此以1980年为界将所有赛季分为两个阶段,分别针对不同阶段赛季的数据进行建模。最后基于10种目前最常用或最流行的机器学习算法,以“高级数据”中的变量为预测变量,以MVP“得票率”为目标变量,利用R语言中的Caret包构建3个集成模型,从中挑选出整体预测精度最好的cubist算法,根据最新的2018-19赛季数据预测NBA常规赛MVP的获奖概率:“字母哥”扬尼斯·阿德托昆博将小胜詹姆斯·哈登获得MVP。Through machine learning,the MVP voting of the NBA regular season was analyzed using the statistical data of the NBA regular season from 1956 to 2018.The linear correlation and the importance of variables were used to group and reduce the dimensionality of technical statistical variables of players in NBA regular season.It was found that the variables in the“complex data”have significant correlations with“Share”,and the“WS”has the greatest impact.At the same time,it was found that since the MVP voting system changed significantly around 1980,the relationship between“Share”and variables also changed significantly.Therefore,all seasons were divided into two stages with the year of 1980,and the data of different seasons were modeled separately.Finally based on 10 of the most popular machine learning algorithms,the variables in the“complex data”were used as predictors,the“Share”was used as the target variable to build three ensemble models with the Caret package in R language.The cubist algorithm with the best overall prediction accuracy was selected to predict the winning probability of NBA MVP based on the 2018-19 season data:Giannis Antetokounmpo will edge out James Harden to get the MVP.

关 键 词：NBA MVP 机器学习 数据挖掘 R语言 

分 类 号：G841[文化科学—体育训练] TP181[文化科学—体育学]