Prediction of Injuries in CrossFit Training: A Machine Learning Perspective

Round 1

Reviewer 1 Report

In this study, the authors proposed a machine learning model to predict injuries in CrossFit training. They assessed different models and found that AdaBoost may be suitable for this problem. Some major points should be addressed as follows:

1. There must have some external validation datasets.

2. Table 1 should be shown with some statistical tests and p-values.

3. The authors should describe in more detail on hyperparameter tuning of the models.

4. Uncertainties of models should be reported.

5. When comparing the predictive performance among models, the authors should conduct some statistical tests to see the significant differences. Then it will be confident to pick up the ones with significant improvements compared to the others.

6. The authors should compare the predictive performance to previously published works on the same dataset/problem.

7. Measurement metrics (i.e., sensitivity, specificity, ...) have been used in previous studies such as PMID: 34502160, PMID: 34989149. Thus, the authors are suggested to refer to more works in this description to attract a broader readership.

8. Source codes should be released to replicate the study.

9. Table 2 is common and not necessary.

10. Besides ROC curve, the authors should also show the PR curve.

11. There must have spaces before reference number.

12. The authors should have a definition for CrossFit training. It will be useful for people in a variety of fields.

13. English writing should be improved.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

The paper presents a methodology for identification risk factors, which are important for crossfit injuries. ML models are created and evaluated with predictive purposes. The data are gathered after surveys. The obtained results are discussed. 

I recommend the authors to check the template. For example, the literature sources should be placed in [], but not in (). After the figure caption should be placed dot (Figure 1., but not Figure 1). Also, should check for typos. 

I suggest, the authors to explain how the formula (1) is derived - empirically or theoretically.

I suggest "using Ensemble Learning" from the title to be removed, because some ML algorithms are used in these experiments.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

In this paper, the authors identified risk factors in CrossFit and developed machine learning-based models based on ensemble learning to predict CrossFit injuries. They collected the relevant data from CrossFit practitioners in Greece and used these data as input data of machine learning algorithms. The experiments demonstrated that their best model could achieve an area under the curve (AUC) of 77.93%. The detailed comments on the manuscript are as follows.

1. The authors should follow the reference format.
2. The authors did not present the details of the questionnaire used to collect data in this manuscript. Even though the authors stated that their questionnaire was based on the previous studies, they also mentioned that they modified it. Thus, the descriptions of the questionnaire are required.
3. There is no background of equation (1). At least, it is necessary to add a reference related to the equation.
4. Generally, typical ensemble models are random forest (or bagging) and boosting models such as gradient boosting machine and AdaBoost. Why did the authors not apply random forest to this study?
5. Some abbreviations (e.g. FS) were used without their definitions.
6. It is needed to improve the readability of Table 3.
7. Figure 1 should be revised. The current form of the figure was not effective in illustrating TP, FP, TN, and FN, indicated by red and green regions. In other words, gray regions should be smaller than the aforementioned regions.
8. More experiments are required to demonstrate the effectiveness of the proposed method. Evaluating the performance differences caused by feature selection can be one of the additional experiments.
9. There is no detail of the experimental settings, such as the list of preselected hyperparameters, the experimental tools, and why the authors chose six factors.
10. Errors

• Line 61, “Sport injuries have significant impact in the national health systems”
• Lines 62-64, “injuries from such as demanding exercise programme like CF may result in loss of man-power, increased costs due to extensive medical treatments and rehabilitation as well as poor quality of life.”
• Lines 140-141, “Partici-pants who had an injury during CF and the injury was confirmed”
• Line 225, “AUC stands for "Area under the ROC Curve”
• Table 3, “26,84”

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

My previous comments have been addressed.

Reviewer 3 Report

The authors sufficiently handled my concerns.