The investigation was aimed at comparing regression and neural models with respect to their accuracy of predicting sports results (Murakami et al., 2005). The presented study involved a group -Wilk normality test and by the homogeneity test (Leaven criterion). All variables presented normal distribution and homogeneity. Correlation matrix and analysis of regression revealed four predictors (cross step, specific power of the arms and the trunk, specific power of the abdominal muscles and the grip power).Then, non-linear regression models as well as neural models were built (Maier et al., 2000). Thus, to verify models, the sports results were predicted for the group of 20 javelin throwers from the Polish National Team (age 18-19 years old) in May 2010 and tested in May 2011 by comparing the models-generated predictions (May 2010) with the actual results achieved by the same javelin throwers (average of three throws after 30 min. warm up). The non-linear regression models (R2=0.871) and perceptron networks structured as 4-3-1 (NRMSE- learning:0.224, validation:0.233 and test series:0.218), demonstrated their capacity for making generalization and predicting sports results. What`s more the difference in the absolute errors values was 10.54m (between true and estimated performances in verification group of 20 Polish javelin throwers), favoring the neural models. The neural model had better goodness of fit for athletes achieving medium or weak results. The negative total error of the network indicates that the model makes larger errors in athletes who throw the javelin further. The analysis of the above data clearly shows that the neural model better predicts sports results than the regression model, confirming also the Bartlett et al. findings (1996), whose neural models provided predictions of better quality than the multiple regression models. Murakami et al. (2005) indirectly proved that neural models are capable of better predictions than nonlinear or linear regression models. Therefore, the investigation demonstrated a significantly greater accuracy of prediction for the perceptron models.
© Copyright 2012 World Congress of Performance Analysis of Sport IX. Veröffentlicht von University of Worcester. Alle Rechte vorbehalten.