INTRODUCTION: Neural networks are able to predict the future success of talents by revealing distinct patterns in the individual setup of the sport specific disposition (Philippaerts, Coutts & Vaeyans, 2008). The purpose of this paper is to compare linear and nonlinear talent prognoses in the crawl sprint. METHODS: The Magdeburg Talent Study on Elite Sport Schools (MATASS) is a six year longitudinal study. The data were collected from 1997 to 2001 from a total of 729 male (age: M = 171.2 months, SD = 42.5) and female swimmers (age: M = 159.3 months, SD = 39.0). The final competition performance data were recorded in 2006 for all male (n = 130) and female swimmers (n = 113). RESULTS: 33 performance prerequisites were measured at three different time points, and reduced by factor analyses: (1) body stature, (2) maximum and explosive strength, (3) general and (4) sport specific speed strength, (5) technique and coordination, and (6) elementary speed. In a second step, the factor values of the six juvenile talent criteria, together with the (7) speed of performance development, (8) utilization of performance prerequisites, and (9) psychological stress stability were used to predict three final talent groups at adult age. For the cross-validated prognosis two methods were used: a linear discriminant analysis (DA), and a nonlinear operating Self-organizing Kohonen Feature Map (SOFM). The comparison of the real adult performance groups with the modeled outcome led to far better predictions in the SOFM. The percentages of correctly predicted cases (females = 87.9 percent; males = 68.3 percent) are much higher than those delivered by the DA (females = 69.0 percent; males = 50.0 percent). DISCUSSION: The quality of the predictions of the linear DA was comparably lower than that of the nonlinear SOFM. Thus, the results of the study show that neural networks are excellent tools to model and to predict future competitive performances on the basis of juvenile talent makeup. Besides that, the better results of the neural network modeling support the interpretation that the development of talented young athletes is a nonlinear complex problem that should be addressed by a dynamic systems approach.
© Copyright 2010 Biomechanics and Medicine in Swimming XI. Veröffentlicht von Norwegian School of Sport Sciences. Alle Rechte vorbehalten.