Author(s): LOWE, G., GOTTWALD, V., JONES, E., OWEN, J., Institution: BANGOR UNIVERSITY, Country: UNITED KINGDOM, Abstract-ID: 483

INTRODUCTION:
The constraints faced by sports organisations due to limited resources reinforce the need for effective talent identification (TID) and selection protocols. Despite previous studies revealing the complex, dynamic, and multidisciplinary nature of TID, there is a prevailing tendency to concentrate on cross-sectional, linear, and unidimensional methods. Subsequently, it is not surprising that practitioners and academics have a poor track record of successful TID systems. More recent scientific approaches have begun to adopt more sophisticated machine learning techniques, which allow for a much more comprehensive exploration of the multifaceted factors likely involved in the identification of talent. Nevertheless, research investigating important determinants of expertise for athletes performing at the highest levels in the UK high-performance systems has adopted what is an arguably flawed ‘one-size-fits-all’ approach. However, there is robust evidence supporting the notion that factors influencing expertise development are almost certainly sport specific. Thus, the purpose of this two-fold study was to apply a machine learning approach to examine the factors influencing progression and success within the Welsh Rugby Union (WRU) pathway.
METHODS:
Attributes from five distinct themes based on underpinning theoretical hypotheses (i.e., anthropometrical, physical, psychological, socio-demographic, skill acquisition, technical and tactical) were assessed across five age-grades during a 3-year collection. An athlete psychosocial survey was employed to measure constructs of behaviours, personality traits, and psychological factors. Overall, 41 constructs were measured. A socio-demographic and skill acquisition questionnaire were used to obtain information about relative age, place of development, family information, education, practice and training factors and sport history. Physical testing included anthropometrical measures, grip strength, isometric mid-thigh pull, countermovement jump, 10-40m sprint, 505 agility, and bronco test. Players’ technical and tactical ability were collected from the coaches using a bespoke tech-tac measure. The study adopted machine learning analysis by way of increasing predictive power. The analysis was used to analyse large numbers of features and find which features best distinguish between two different classes of objects, in relation to progression and success.
RESULTS:
Based on over 87 variables, we were able to identify key features across psychosocial, physiological, and skill acquisition-related domains that could be used to discriminate progression and success within the WRU pathway.
CONCLUSION:
Despite recent encouragement of multidimensional approaches in TID, many sporting organisations still exhibit a bias towards physical and anthropometrical factors. The model can help support the efficaciousness of talent systems and assist key stakeholders in implementing identification and selection protocols.