& 150 APPENDICES advanced machine learning algorithms to predict whether they would achieve their step goals. By automatically analysing physical activity and physical performance, timely detection of anomalies in behaviour and identifying effective coaching strategies may become possible. The results showed that tree-based algorithms best predict whether an employee will achieve his or her step goal. In Chapter 3, we studied the predictability of physical performance in elite soccer matches using various physical performance measures and machine learning techniques. We gathered data from 302 matches in a single season using the SportsVU optical tracking system, which recorded the positions of each player throughout the matches. Based on this data, we measured physical performance using three increasingly sensitive performance measures, i.e. distance covered, distance in speed zones, and energy expenditure in power zones. These physical performance measures were used in different machine learning models to identify and predict the physical performance of individual players throughout an elite soccer match. We found that the more sensitive the performance measures were, the better the physical performance of the individual player could be predicted. In Chapter 4, we studied the impact of substitutions on a soccer team’s physical performance using a causal roadmap and causal model. The causal roadmap strictly prescribed which steps should be taken in analysis and clarified the underlying assumptions. In which a causal model provided insight into how reality is modelled, and which variables have been left out. Our causal model included variables such as the number and timing of substitutions and the total distance covered. The causal roadmap and causal model helped us identify assumptions and potential sources of bias and confounding that could affect causal effect estimates. In Chapter 4, we also provided an in-depth analysis of statistical methods. We tested the accuracy of estimating the impact of substitutes on a football team’s physical performance using replacement data and data from the SportsVU optical tracking system. We compared the accuracy of two methods: Targeted Maximum Likelihood Estimation (TMLE) and a generalised linear model based on the complete dataset. We also tested the accuracy of these methods when a critical variable was removed from the dataset. The more robust TMLE method offered more accurate insights than the generalised linear model, especially in the absence of a critical variable in the dataset. In Chapter 5, we examined the impact of workload on injuries in runners by applying both statistical analysis and machine learning. Our dataset consisted of individual data from test, training sessions, and injury logs. We used physical load measures like training duration and rate of perceived exertion to construct an acute:chronic workload ratio. Our
RkJQdWJsaXNoZXIy MjY0ODMw