Learning Performance Efficiency of College Basketball Players Using TVAE

Monitoring workload is essential to evaluate possibility of fatigue or injuries and overall performance of the players. Large player’s workload in any game including basketball can contribute to the stress and fatigue and overloaded players may get exhausted and exhibit burnout symptoms, which eventually result in their lower efficiency in the game. This paper aims at predicting the efficiency of the basketball players in all positions (guard, forward and center) based on their workload information. Machine learning (ML) methods for regression and classification are applied to the dataset for predictive modelling of the variables. The analysis includes: (i) one-model for all player positions and (ii) position-based models for respective player position. Leveraging tabular variational autoencoders (TVAE), synthetic data is generated to improve the accuracy. The evaluated models from both regression and classification verify that better accuracies can be obtained in the position-based models rather than the one-model for all positions approach. The performance analysis of the algorithms indicates that the player’s efficiency can be estimated from the workload information which can provide valuable insight and individualized recommendation for optimal performance during the competition.