Van het ongrijpbare naar het meetbare: Machine learning voor het modelleren van prestatiebepalende factoren in de sport

Over the past two decades, sports analytics has emerged as a powerful tool for advancing our understanding of athletic performance and ultimately gaining a competitive advantage in sports. The use of wearable sensors, video tracking, and other monitoring technologies has enabled large-scale data collection. When combined with statistical analysis, these data have transformed how teams and athletes approach training, competition, and strategic decision-making. Data-driven insights now underpin talent identification, opponent scouting, injury prevention, and load management. Despite these advancements, certain aspects of sports performance remain difficult to quantify. Many of the qualities practitioners seek to understand—such as leadership, teamwork, and decision-making under pressure—are inherently intangible and cannot be directly measured through objective data points. Other parameters may be too costly or impractical to assess regularly. For example, in endurance sports, laboratory tests such as VO₂max protocols and force-plate assessments provide gold-standard measurements but are impractical for frequent, in-field use. This dissertation explores how machine learning can provide value to the analysis and interpretation of these elusive aspects of performance. The work focuses on two specific domains: (1) the analysis of biomechanical parameters in outdoor running, and (2) the assessment of in-game decision-making in professional soccer. In running, monitoring biomechanical load is essential for injury prevention, but obtaining such data typically requires expensive, lab-based equipment. In this dissertation, we explore how machine learning can provide a scalable alternative by estimating key biomechanical parameters from wearable sensor data. Specifically, we use bilateral shin-mounted accelerometers to estimate (1) gait events to derive ground contact time, and (2) vertical instantaneous loading rate (VILR), a kinetic variable linked to injury risk. Our results demonstrate that machine learning models can accurately estimate these parameters during overground running and outperform heuristic methods, offering a promising step toward real-time, field-based feedback systems for injury monitoring and prevention. In soccer, decision-making is central to performance, yet data only captures the action taken—not the alternatives a player could have considered—making it difficult to assess whether a player made the optimal choice in a given situation. This dissertation approaches decision-making from three complementary angles. First, we examine how context influences decisions by modeling both pre-game and in-game factors to estimate mental pressure, allowing us to assess how players respond under varying psychological demands. Second, we explore creativity as a specific aspect of decision-making, identifying effective passes that deviate from typical choices. Third, we focus on pressing as a tactic that requires coordinated decision-making and propose a framework to quantify the impact of pressing in terms of how a decision to press affects the long-term risk and reward. Together, these frameworks offer a data-driven, objective lens on decision-making in professional soccer. Together, these case studies illustrate the promises and limitations of machine learning to assess complex, difficult-to-measure aspects of sports performance. While our models deliver actionable insights, their validity hinges on careful feature design, critical evaluation, and close collaboration with coaches and sport scientists. Future research should prioritize athlete- and team-specific modeling to capture individual style and tactics, and integrate richer contextual variables such as fatigue, opposition behavior, and environmental factors. Advances in interpretable modeling and uncertainty quantification will be essential to foster practitioner trust and ensure that analytics guide, rather than oversimplify, decision-making in high-stakes sporting environments.