A Mixed Model for Performance-Based Classification of NBA Players

Abstract

Using data collected from the Basketball-Reference.com, this study examined NBA player performance values to discern patterns and to classify clusters exhibiting common patterns of player performance. Empirical results based on the K-means clustering analysis identified three NBA player clusters. Results of the K-means clustering analysis were tested for accuracy using the discriminant analysis indicated that cluster means were significantly different. The results of one-way ANOVA also showed that significant differences in all twenty-one independent variables were found within the three identified NBA player clusters. The multilayer perceptron neural network model was utilized as a predictive model in deciding the classification of NBA players based on their performance related statistics. From an architectural perspective, it showed a 21-7-3 neural network construction. Results of this study may provide insight into the understanding of the performance of NBA players for NBA management purposes.