On predicting an NBA game outcome from half-time statistics

Abstract

Abstract Predicting the outcome of an NBA game is a major concern for betting companies and individuals who are willing to bet. We attack this task by employing various advanced machine learning algorithms and techniques, utilizing simple half-time statistics from both teams. Data collected from 3 seasons, from 2020/21 up to 2022/23 were used to assess the predictive performance of the algorithms at two axes. For each season separately, apply the algorithms and estimate the outcomes of the games of the same season and secondly, apply the algorithms in one season and estimate the outcomes of the games in the next season. The results showed high levels of accuracy as measured by the area under the curve. The analysis was repeated after performing variable selection using a non-linear algorithm that selected the most important half-time statistics, while retaining the predictive performance at high levels of accuracy.

Keywords

• NBA
• Half-time statistics
• Game outcome
• Machine learning