Optimising the Use of Machine Learning and Computer Vision in Sport: An Ecological Dynamics Perspective.

Although machine learning and computer vision is a growing area of research in sports analysis, its implementation in athlete development programs does not guarantee performance improvements. Developers typically design and implement machine learning and computer vision technologies into athlete development programs because of the high level of technical information on performance that can emerge. The value gained from this approach can be limited by siloed working practices in sports organizations and by sporadic approaches to athlete development which can negatively affect skill development. Here, we discuss why the design and integration of machine learning and computer vision in athlete development programs needs to be rationalized by a theoretical framework to guide effective collaborations between sport scientists, technologists, and practitioners. This position paper illustrates how the use (i.e., design and implementation) of machine learning and computer vision technologies in athlete development programs could be underpinned by the structural organization of a Department of Methodology (DoM), and that underpinned by a theoretical framework, such as ecological dynamics. We outline how the integration of machine learning and computer vision technology, underpinned by an ecological theoretical approach, can accomplish the following: (1) support representative learning design, (2) individualize training and assessment of athletes, and (3), enhance, but not replace, the quality of coaching within athlete development programs.