NBA Playoff Prediction Based On Machine Learning Algorith

The influence of sports events in people’s daily life is expanding and various sports events are emerging one after another.Among the many sports events,the NBA is a sports league with an earlier development,greater influence and stronger brand influence.There is a large amount of game data recorded in the NBA’s daily games,and the research and utilization of these data has great commercial value,so it has become an important topic.Among the many research topics,playoff prediction is an important aspect.The playoffs are divided into four rounds,which are the first round of the playoffs,the division semifinals,the division finals and the NBA Finals.The two teams in the match adopted a four win system in seven sets.Winning the championship and entering the playoffs is usually the goal pursued by teams and players.For the team,entering the next round of games can get more ticket revenue and a lot of exposure,which has great commercial value.The results predicted by the model can guide the team to sign new players and arrange tactics more reasonably to win.At the same time,the idea of building the model can also be applied to other sports fields.In this paper,two types of comprehensive data for the first round of the playoffs and the last three rounds of the playoffs are established based on game data from the regular season,historical data from the coach,comprehensive evaluation data from the players,and historical data from both teams.Based on Person correlation analysis and variable importance analysis,some key variables that affect the outcome at different stages are obtained.Semi-supervised model and supervised model are respectively established on the two types of established data.The semi-supervised model is mainly unsupervised clustering,and the supervised model include logistic regression with penalty,support vector machine,decision tree,naive Bayes and random forest.The semi-supervised clustering model established in the first stage has achieved good results in the prediction process.The semi-supervised clustering model is obtained on the basis of the clustering model by adding "must-link" and "cannot-link " constraints.Semi-supervised clustering reasonably utilizes the historical statistics of the first round of the playoffs to give the winning and losing labels of the classified teams.Logistic regression with penalty,support vector machine,naive Bayes and random forest algorithms are used for the two kinds of data of the division semi-finals,division finals and NBA finals respectively,and the prediction data is established based on the performance of the team in the regular season.The prediction effect of naive Bayes on historical data is better,and the prediction effect of logistic regression on comparative data obtained by comprehensive evaluation data is better.In this paper,the playoff predictions are organized and modeled in stages to obtain more accurate prediction results.