Institute of Sports Science

Talent identification and development in football

As part of a larger research program on talent identification and development in football, this project within the Data Science and Sports Lab focusses on the application of machine learning techniques and Bayesian modelling (dynamic latent variable approaches) for the prediction of future success based on complex data gained from talent assessments. In a first study, we evaluate the utility of machine learning techniques (e.g., neural networks, boosting methods, and support vector machines) for examining prognostic validity of speed abilities and technical skills in youth elite football.

Research Line: Athletes, Teams and Performance.

Funding / Support: German Football Association (Deutscher Fußball-Bund e.V.).

Project Team: Oliver Höner, Augustin Kelava, Pascal Kilian, Daniel Leyhr

Publications:

  • A deep learning factor analysis model based on importance-weighted variational inference and normalizing flow priors: Evaluation within a set of multidimensional performance assessments in youth elite soccer players. Published in the Statistical Analysis and Data Mining: The ASA Data Science Journal.
  • Soccer goalkeeper expertise identification based on eye movements. Published in PloS one.
  • Expertise Classification of Soccer Goalkeepers in Highly Dynamic Decision Tasks: A Deep Learning Approach for Temporal and Spatial Feature Recognition of Fixation Image Patch Sequences. Published in Frontiers in Sports and Active Living.