This paper uses NBA 2014 - 2015 regular season as one of the examples of competitive sports and tries to find a general model for the prediction of other competitive sports. Firstly, considering the influence of the game results and scores in the home court and on the road, we fixed the weight of the wining of the game and defined the weight of the scores in the home court and on the road. Then we could get the prior probability of the wining probability between each team. Secondly, we used the principal component analytical method to reduce the dimension of the technical statistics we have collected. The total number of dimension of the principal components is 50. That means each team is described by the 50 variables. Next, we used stepwise regression, multiple linear regression, quadratic polynomial regression and Logistic regression model to make prediction of the results of the games. We found that it is hard to make the average deviation under 10%. Then we use BP artificial neural network model to make the prediction. After 5000 time iterative learning, the deviation could be reduced under 10%. Finally, we use Adaboost algorithm to combine the classifiers. We use 20 BP neural networks as the weak classifiers. After the adjustment of the number of nodes in hidden layer, we could make the average deviation under 5%. In the meantime, we also discussed the relationship between the number of nodes in hidden layer and the performance of the Adaboost BP network. Comprehensively considering the promotion that Adaboost algorithm brings about and the time complexity of the networks, we define a standard to describe the property of the networks. We found out that the network performs the best when the nodes in hidden layer is 28.

Creative Commons License

Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.