Bergsma, B. (Berber) (2021) Putting machine learning into practice: predicting readiness to train in elite short track speed skaters. thesis, Sport Sciences.
Full text available on request.Abstract
Background: Monitoring training load and recovery is a key element in the design of training programs for elite athletes. Having insight in athletes’ readiness to train (RTT) supports coaches in designing training programs that are tailored to the individual athlete (Gabbett et al., 2017). Machine learning may contribute to understanding such day-to-day monitoring of elite athletes. Aim: The present study aimed to evaluate the predictive capacity of perceptual wellbeing and internal training load measures on overall, physical, and mental RTT in elite short track speed skaters. Methods: Training data were collected from 20 world class short track speed skaters during 31 weeks of the 2020-2021 season. Training load data, perceptual wellbeing scores, and RTT scores were obtained using daily self-report questionnaires. The collected scores were pre-processed into 170 independent variables for a total of 3794 observations. Predictive models were constructed using two machine learning techniques: linear Elastic Net regression and non-linear Decision Tree regression. Overall, physical, and mental RTT scores were predicted over a 1, 3 and 7 day time frame. Model performance was evaluated using the mean absolute error and explained variance. Additionally, model performance was compared with the naive baseline model performance using paired sample t-tests. Results: Both the linear (p < 0.001, d = 0.15) and non-linear model (p < 0.001, d = 0.39) outperformed the baseline model in predicting overall RTT over the 1-day time frame. However, the non-linear model showed higher accuracy for physical (p < 0.001, d = 0.40) and mental (p < 0.001, d = 0.47) RTT prediction compared with the linear model. In general, prior RTT scores over the 3-day time frame were found to be important variables in RTT prediction. The machine learning techniques used for model constructed did not outperform the baseline model in predictions of further ahead RTT scores. Conclusions: Machine learning techniques, and non-linear regression modelling in particular, may have added value in predicting RTT scores one day ahead. It contributes to the optimalization of training program design for high intensity sport athletes like short track speed skaters. Additionally, Decision Tree regression modelling may support practitioners in selecting important variables in this monitoring process. Keywords: athlete monitoring, predictive modelling, readiness to train, perceived wellbeing, training load
| Item Type: | Thesis (UNSPECIFIED) |
|---|---|
| Supervisor name: | Kempe, dr. M. |
| Faculty: | Medical Sciences |
| Date Deposited: | 11 May 2022 10:22 |
| Last Modified: | 11 May 2022 10:22 |
| URI: | https://umcg.studenttheses.ub.rug.nl/id/eprint/3286 |
Actions (login required)
 |
View Item |