Investigating Machine Learning Applications in the Prediction of Occupational Injuries in South African National Parks

Reviewer 1

This paper was intended to apply various machine learning classification models to predict occupational injuries in parks and nature reserves. I think overall the organization of this paper is OK, but I do have some comments that the authors might find useful to further improve the quality of this study:

Review

Changes in the Article

1) For the injury classification, I do not think there are no injury levels associated with each class, a note under the table will make this clear to the readers;

A note has been made under Table 2: Response Variable which is under subsection 2.3 Definition of Categories of the Response Variable that is in Section 2: Materials and Methods.   

2) the other three variables do not seem to highly correlate with the injury classification, the injury classification is self-explanatory; not sure we need these many variables in the Tables, if not significant, it is suggested to remove them;

Table 1. under section 2.2 Definition of features in the Materials and Methods section.

3) In Figure 1, I think the description of the injury classification besides the code is better to distinguish from each other;

Description of the injury codes are given under Figure 1.

4) entropy, Gini index and Classification error are used, can the authors provide more details about those indicators, such as equations;

These were not used in the models explored and therefore any description of these has been removed.

5) 4.3 Prediction of Injuries for employees with lower ... I think section 4.2 is missing.

Under the results section adjustments were made such that there are subsections 3.1 Prediction of Injuries for employees with lower extremity injuries and injuries in the Torso and hands region and 3.2 Importance of Features.

Reviewer 2

The manuscript is aimed to interesting topic regarding testing different machine learning classification models for prediction of occupational injuries in parks and nature reserves. The paper requires a minor revision and corrections. I have following questions and suggestions for the authors:

Review

Changes in the Article

In abstract section I suggest that the authors indicate the country where the research was conducted

South African National Parks is mentioned in the abstract section.

L36: Also, for key words I suggest smaller number! Perhaps omit the names of the used models and include protected natural areas, classification models etc

Keywords: Machine learning, Prediction, Occupational Injuries, National Parks

L52-53: Usually when writing scientific paper, the fundamental goal is stated at the end of the introduction, which is not the case in this paper! I suggest that the authors correct that.

In introduction chapter, did I miss disposition paragraph?

Under the introduction section the main aim and the purpose are mentioned as follows:

Therefore, the main aim of this research study is to investigate the application of machine learning in the prediction of occupational injuries using underlying factors such as demographic, period which the accident occurred and injury related factors. Several machine learning models are appropriately applied to real world data and their performance in the prediction of occupational injuries are critically assessed. The purpose is to investigate whether the best performing model in the prediction of injuries can be used to identify employees who are at risk of occupational injuries and consequently develop timeous interventions and prevention measures. 

L79: The ordinal number in the chapter on materials and method is number 2, not 3! I suggest the authors to correct the chapter numbering throughout the paper!

The section is now 2. Materials and Methods in accordance with the guidelines from the MAKE Microsoft template.

L96; L110: put the space between the table and the picture and the number associated with it!

Noted and changes have been made accordingly to the recommendations for presenting tables and figures.

L250 and L279: Table 3 and Table 5 and accompanying text is more suitable for chapter methods and materials! I suggest the authors to make the necessary changes!

Moved to the materials and methods section as Table 4 under subsection: 2.6 Prediction of Injuries for employees with lower extremity injuries and injuries in the Torso and hands region.

L257 and L288: I suggest the authors for table 4 and 6 to keep the data related to the classification table by model and to delete Confusion Matrix from the paper (if it is necessary to display the Confusion Matrix, the authors are suggested to present only the most important ones in the form of a figure and link them to the accompanying text).

This was corrected in the results sections. Table 5 presents the results and confusion matrices are removed

Why were the same input parameters not used in the model settings in the injury analysis in subsections 4.1 and 4.2? Please explain how this affects the results obtained and their comparison!

There is only Table 4 in the reviewed article which shows the parameters of the models for the prediction of lower extremity injuries and injuries in the Torso and hands region.

I would like to see more precise conclusions, evidently matching your purpose (a) clearly related to the overall results of the whole sample and comparation to similar studies, (b) clearly related to the results from subsections 4.1 and (c) clearly related to the results from subsections 4.2.

Under the Discussion Section these are the following findings based on the purpose which was outlined in the last paragraph of the discussion chapter:

·       The overall accuracy of the SVM model performed well below the findings of other previous studies that have used SVM models to predict occupational injuries or accidents and reported accuracy levels above 90%. [7,8].

·       Furthermore, the performance was below the accuracy of other previous studies that had used Neural Networks in the prediction of work-related injuries that had achieved accuracy levels greater than 80% [4, 8].

·       Furthermore, gender was the only feature that had importance scores that were significantly greater than zero. Findings in previous literature have found that the rate of claims being made for occupational injuries have been found to be higher for males than females [16]. This indicates that the gender of the employee can potentially influence the chances of experiencing occupational injuries.

·       The application of the findings in this research study are only limited to the prediction of whether employees who would have previously experienced injuries in the lower extremity or in the Torso and hands region will experience any of these injuries again because the data was only filtered for these types of injuries. This will be useful in the identification of employees of who are vulnerable to experiencing these injuries again in the future. The organisation can then implement targeted preventive or safety measures and provide safety training for these vulnerable employees.

Reviewer 3

The Authors deal with an interesting topic and the present paper describes very important issue. The topic is of interest to safety professionals as well as academic. However, the paper is poorly written and substantial changes need to be made in order for the paper to be considered for publication. My main issue is that the structure of the paper is very poor and does not represent a scientific research paper.

Review

Changes in the Article

1.       The title is interesting, although it may be clearly indicate the sector of the economy - occupational injuries in the Parks and Nature reserves.

Noted the title has been changed to the following” Investigating Machine Learning applications in the prediction of occupational injuries in South African National Parks”.

2.       The abstract is too long. According to the "Instructions for Authors":

Abstract: The abstract should be a total of about 200 words maximum. The abstract should be a single paragraph and should follow the style of structured abstracts, but without headings: 1) Background: Place the question addressed in a broad context and highlight the purpose of the study; 2) Methods: Describe briefly the main methods or treatments applied. Include any relevant preregistration numbers, and species and strains of any animals used. 3) Results: Summarize the article's main findings; and 4) Conclusion: Indicate the main conclusions or interpretations. The abstract should be an objective representation of the article: it must not contain results which are not presented and substantiated in the main text and should not exaggerate the main conclusions.

The abstract has been changed accordingly under the abstract section. It has 197 words.

3.       Keywords is too words - max 5 words

Keywords: Machine learning, Prediction, Occupational Injuries, National Parks   

There are only four keywords used

4.       Authors should use the Microsoft Word template or LaTeX template to prepare their manuscript proposed by the Journal. (MAKE | Instructions for Authors (mdpi.com)

The whole document has been changed according to the Microsoft word template accordingly

5.       The introduction (Section 1) describe the research background very poor. The articles cited are not known in the international literature, they have no citations. There are many more significant papers on this topic and the literature review should be improved

The introduction section has been changed accordingly. Literature relating to the prediction of occupational injuries includes the following journal articles:

·       5. Cheng, C. W. et al., 2011. Applying data mining techniques to explore factors contributing to occupational injuries in Taiwan’s construction industry. Accident Analysis and Prevention, Volume 48, pp. 214-222.

This article has been cited 187 times and is from recognized journal

·       6. Debnath, J. et al., 2016. Fuzzy inference model for assessing occupational risks in construction sites. International Journal of Industrial Ergonomics, Volume 55, pp. 114-128.

This article has been cited 54 times and is from an international journal

·       7. Sánchez, A. S. et al., 2011. Prediction of work-related accidents according to working conditions using support vector machines. Applied Mathematics and Computation, Volume 2018, pp. 3539-3552.

This article is from a well-recognized and reputable journal. The Impact Factor of this journal is 4.397, ranking it 7 out of 267 in Mathematics, Applied. The journal is indexed in 12 international databases

·       8. Sarkar, S. et al., 2019. Application of optimized machine learning techniques for prediction of occupational accidents. Computers and Operations Research, Volume 106, pp. 210-224

This article has been cited 119  times and is from a  recognized  journal

·       4. Ivaz, J. et al., 2021. Prediction of the Work-related Injuries Based on Neural Networks. CzOTO, 3(1), pp. 19-37

The articles are from recognized journals which have been cited.

6.       I do not see Section 2 ???

Section 2 is Materials and Methods according to the Microsoft word template from (MAKE | Instructions for Authors (mdpi.com)

7.       The Section 3 (Materials and Methods) should be dedicated to describing this methodology and what you did in your paper. The methodology should be described and be solid enough such that any other person using the same procedure will could repeat the research. Now, it is impossible. Now this section contains only short description of the method and nothing else.

Materials and Methods according to the Microsoft word template from (MAKE | Instructions for Authors (mdpi.com) is now Section 2 in the reviewed article.  This section has been broadened to explain the following:

2.1 Acquisition of Data

2.2 Definition of features

2.3 Definition of Categories of the Response Variable

2.4 Machine Learning Models

2.5 Evaluation

2.6 Prediction of Injuries for employees with lower extremity injuries and injuries in the Torso and hands region