6 138 CHAPTER 6 of physical performance measures. The literature [4], [5] has acknowledged that the accuracy of an algorithm can differ across datasets due to various factors, including feature selection, such as physical performance measures. Therefore, it is crucial to choose the physical performance measures carefully. The use of various machine learning algorithms influences the quality of predictions positively. Chapter 2 involved training eight machine learning algorithms and comparing their performance to a baseline algorithm. The authors determined the optimal algorithm for a given dataset by assessing the performance of each algorithm, with the Random Forest algorithm demonstrating superior predictive ability for individual thresholds, while the ADABoost algorithm showed the highest precision for group-level predictions, which aligned with previous research findings [4], [5]. Furthermore, as discussed in Chapters 2 and 3, we found that the choice of machine learning algorithm significantly impacted the precision of individual predictions, highlighting that determining the most suitable algorithm for a given dataset is a context-dependent art rather than a science. Identifying the best combination of the dataset and algorithm is time-consuming. Identifying the best combination might be accelerated by utilising ensemble methods, combining multiple algorithms to produce one cohesive model [5]. For example, Chapter 4 employed an advanced ensemble method Super Learning. Super Learning is supposed to outperform other single algorithms by automatically selecting the best algorithm or combination of algorithms [6] to assess the impact of a substitute player on a soccer team’s physical performance. However, even when an accurate machine learning model is constructed, machine learning models can become outdated, for machine learning models are confronted with a constantly changing environment and data [7]. For instance, the individual player performance can change over time due to factors such as injuries, fatigue, or changes in form. This can affect the team’s performance and lead to a drift in the machine model’s predictions. Therefore, it is crucial to continuously monitor the accuracy and precision of machine learning models and retrain the machine learning models when their predictions no longer correspond with the changing circumstances and data [7]. Following a causal roadmap in combination with a causal model prevents oversimplified assumptions. By following a causal roadmap and creating a causal model in Chapter 4, depicting the causes and effects of a substitution in soccer, we could identify the factors that
RkJQdWJsaXNoZXIy MjY0ODMw