Position-specific performance profiles, using predictive classification models in senior basketball

Basketball players display different performance characteristics when in different playing positions. Traditional statistical techniques such as Multivariate Analyses of Variance (MANOVA's) are insufficient when predicting specific positions. Alternatively linear statistical models, such as discriminant analysis, have been used. Recently non-linear statistical methods have been introduced into sport science via artificial neural networks that have been proven to have high potential. This study will seek to identify whether artificial neural networks are capable of providing additional insights with regards to the position-specific characteristics found in basketball. A total of 150 Belgian elite players performed physical and physiological tests in the preseason phase. Linear and non-linear predictive models were applied. Discriminant analysis and multi-layer perceptron analysis were able to position, respectively, 92 and 88% of the players correctly. The results of the variable importance analysis demonstrated that the positions clearly differentiated from each other. Herein, weight was the most important factor. Secondly the shuttle run, the speed at anaerobic threshold and the sprint time between 5 and 10 m (respectively, 93.2; 85.0 and 79.5% importance of weight) were important factors. The current study showed that basketball positions clearly differentiate elite Belgian basketball players based solely on basketball independent tests.