[1K5-ES-2] Machine learning: Social application (2)

[1K5-ES-2-01] Predicting the NBA winning percentage base on the linear regression model

〇Wei-De He¹, Yi-Ting Chiang¹, Hung-Jui Chang¹ (1. Chung Yuan Christian University)

キーワード：regression model, stepwise regression, sigmoid function, NBA

This study analyses the NBA regular season result from 2015-19. Traditional statistical data are used as explanatory variables to establish linear regression models. We use the two team's score intervals predicted by ourmodels as the win rate indicator. By using stepwise regression methods to organize the data can eectively improve model accuracy. Experiment results show a 49% of correctness for predict a match with multiple games, and a 92% correctness with at most one game dierence. The main factor of causing incorrect prediction is also recognized as the imbalance competition system.