基于半监督聚类的NBA季后赛第一轮预测
Prediction of the First Round of NBA Playoffs Based on Semi-Supervised Clustering

体育赛事的兴起使得大量的数据被纪录下来，体育统计随之发展起来。在众多体育赛事中，NBA是其中一个影响力较大的体育联盟，在NBA数据的分析中季后赛预测是一个重要的方面。NBA季后赛分为四个阶段，将预测分为多阶段分析建模并进行预测有着现实意义，本文旨在研究季后赛第一轮的预测问题。季后赛的预测实际是一个二分类问题，本文通过整理当前赛季常规赛的比赛统计数据、教练的历史执教数据和球员当前赛季的RPM值，进而从球队、教练、球员三个方面给出球队实力的评价值，并在此基础上建立有勿连约束和必连约束的半监督聚类模型，最后根据历史统计数据给出已分好类的标签，预测结果表明半监督聚类在NBA季后赛第一轮的预测中有着较好的预测效果和很强的适用性。
With the rise of sports events, a large number of data have been recorded, and sports statistics have developed accordingly. Among many sports events, the NBA is one of the most influential sports leagues, the prediction of the playoffs is an important topic in the study of NBA data. The NBA playoffs are divided into four stages. It is of great practical significance to divide the prediction into multiple stages. This paper aims to study the prediction of the first round of the playoffs. The prediction of the playoffs is actually a two-category problem. By arranging the game statistics of the regular season, the historical coaching data of the coach and the RPM value of the players in the current season, we provide the evaluation value of the team’s strength from the three aspects of the team, the coach and the players. On the basis of establishing data, a semi-supervised model with must-link and cannot-link constraints is established. Finally, according to the historical statistical data, the well-classified labels are given. The prediction results show that semi-supervised clustering has a good prediction effect and strong applicability in the prediction of the first round of the NBA playoffs.