Modeling Job Satisfaction of Peruvian Basic Education Teachers Using Machine Learning Techniques
Round 1
Reviewer 1 Report
Many of the figures are too small to read (for example figures 1, 2, 3, ). Please make them larger.
Results: Before stating that one result is "better" or "worse" than another it is important to determine if the difference is significantly different.
Equation derivations can be included in an appendix. (for example Robust Scaling data).
Figure 5 is a table.
Author Response
Response to Reviewer 1 Comments
Point 1: Many of the figures are too small to read (for example figures 1, 2, 3, ). Please make them larger.
Response 1: We accept the suggestion. In fact, we have resized all the figures so that now they can be better appreciated.
Point 2: Results: Before stating that one result is "better" or "worse" than another it is important to determine if the difference is significantly different.
Response 2: You are right, we accept the suggestion. In fact, in order to determine the significance of the differences we have carried out two statistical tests, the same ones that can be seen in Table 6. ANOVA test in Cohen's Kappa metric and Table 7. Duncan's test in Cohe's Kappa metric.
Point 3: Equation derivations can be included in an appendix. (for example Robust Scaling data).
Response 3: We accept the suggestion. Equations derivations are now part of Appendix A.3.
Point 4: Figure 5 is a table.
Response 4: We corrected the bug. We converted it into Table 3.
Reviewer 2 Report
· The authors should include a last paragraph that summarizes the remaining part of the article in the introduction section.
· The author should state the gaps discovered from the literature in the last paragraph in the state-of-the-art section.
· The authors should perform a comparative analysis of their proposed model with state-of-the-art models.
· The dataset size, collection process, and range are not mentioned.
· How did the authors tune the optimal hyperparameter of all models? It should be described clearly.
· Uncertainties of models should be reported.
· When comparing the predictive performance among methods/models, the authors should perform some statistical tests to see significant differences.
· Overall, the English language and presentation style should be improved significantly. There contained a lot of grammatical errors and typos.
· Source codes should be provided for replicating the study.
· The authors should use, cite and reference 2022 and 2023 articles. The most recent articles I could see are published in the year 2021.
Author Response
Response to Reviewer 2 Comments
Point 1: The authors should include a last paragraph that summarizes the remaining part of the article in the introduction section
Response 1: We agree. We add a last paragraph that summarizes the content of the rest of the article, just before section 2. State of the art.
Point 2: The author should state the gaps discovered from the literature in the last paragraph in the state-of-the-art section.
Response 2: We accept the suggestion. We have carried out a more comprehensive review of the state of the art; In addition, we have added a last paragraph in which we establish the scientific gap that we have identified in the study.
Point 3: The authors should perform a comparative analysis of their proposed model with state-of-the-art models.
Response 3: As a result of a better review of the state of the art, which we establish in section 2, now in the results chapter we discuss and make an analysis of our model with the models identified in section 2. We can verify this from line 343 up to line 391 of the pdf.
Point 4: The dataset size, collection process, and range are not mentioned.
Response 4: We believe that what is indicated can be seen in section 3.1 Data cleaning and pre-processing.
Point 5: How did the authors tune the optimal hyperparameter of all models? It should be described clearly.
Response 5: We accept the suggestion. We have now included section 3.3.2. Hyper-parameter tuning and model training
Point 6: Uncertainties of models should be reported.
Response 6: We have carried out an intensive review of articles related to our study and in none of them have we found this indicator, which makes it impossible for us to identify the methodology that we should use to calculate the uncertainties.
Point 7: When comparing the predictive performance among methods/models, the authors should perform some statistical tests to see significant differences.
Response 7: You are right, we accept the suggestion. In fact, in order to determine the significance of the differences we have carried out two statistical tests, the same ones that can be seen in Table 6. ANOVA test in Cohen's Kappa metric and Table 7. Duncan's test in Cohe's Kappa metric
Point 8: Overall, the English language and presentation style should be improved significantly. There contained a lot of grammatical errors and typos.
Response 8: We accept the suggestion. We have conducted a thorough review of the English language writing.
Point 9: Source codes should be provided for replicating the study.
Response 9: The source code is available at the URL that we have included in the Data Availability Statement section:
Point 10: The authors should use, cite and reference 2022 and 2023 articles. The most recent articles I could see are published in the year 2021.
Response 10: We have corrected this. We have included bibliographical references for the indicated years.
Author Response File: Author Response.docx
Reviewer 3 Report
The article is devoted to modeling teachers' job satisfaction in primary education using machine-learning techniques.
This is well-performed research in an important field, but the article has several significant flaws.
1. The discussion of the achieved results is very sketchy. It can be improved in several ways:
1.1. Discussing applicability of the results to teachers in other countries. The study was based on the dataset from a single country (Peru); on which countries its results can be safely generalized? I also recommend changing the article's title to reflect the fact that the study was conducted on Peru teachers.
1.2. How well the study results correspond to the results of previous studies on teacher satisfaction? This will also help to answer the question about generalizability of the results.
1.3. In what ways the study results can be used? This, for example, affects the choice of parameters for selecting the best method as there were no Pareto-optimal method.
2. Linked to the previous problem, the review of Related Work can be improved by reviewing other works in the field of studying teacher job satisfaction so that their results can be compared in the Discussion/Conclusion section. Right now, the review only concentrates on different techniques of predicting job satisfaction in different jobs.
3. The authors correctly state that "Since the data set is partially unbalanced this indicator could be misleading." However, they proceed to select some of these non-reliable indicators instead of using classification accuracy measures that are independent of the imbalance in the dataset (for example, Matthews correlation coefficient, Youden's J or Cohen's kappa). I suggest using indicators that are independent of the number of cases in the classes.
4. The dataset is uploaded to Google Drive which is not permanent storage. I suggest publishing the dataset on the relevant sites and acquiring DOI for it.
Fixing these problems will help to increase the interest and impact of the article.
There are some minor problems in the article such as:
1. Figure 11 is difficult to read because many lines are very close to each other.
2. Variables, listed in Figure 2 and Figure 3, are not described in the article (there is no way to understand their designations from the article text)
3. I don't understand the reason for adding "Memory allocation" to Table 1 because it depends on the way the dataset is represented in the memory.
4. There are minor English problems (e.g., "It is currently known that satisfaction or dissatisfaction {of??} teachers with their work life...")
Author Response
Response to Reviewer 3 Comments
Point 1: The discussion of the achieved results is very sketchy. It can be improved in several ways:
1.1. Discussing applicability of the results to teachers in other countries. The study was based on the dataset from a single country (Peru); on which countries its results can be safely generalized? I also recommend changing the article's title to reflect the fact that the study was conducted on Peru teachers.
1.2. How well the study results correspond to the results of previous studies on teacher satisfaction? This will also help to answer the question about generalizability of the results.
1.3. In what ways the study results can be used? This, for example, affects the choice of parameters for selecting the best method as there were no Pareto-optimal method
Response 1: As a result of a better review of the state of the art, which we establish in section 2, now in the results chapter we discuss and make an analysis of our model with the models identified in section 2. We can verify this from line 343 up to line 391 of the pdf. Likewise, at your suggestion we have changed the title of the article to: Modeling job satisfaction of Peruvian basic education teachers using Machine Learning techniques
Point 2: Linked to the previous problem, the review of Related Work can be improved by reviewing other works in the field of studying teacher job satisfaction so that their results can be compared in the Discussion/Conclusion section. Right now, the review only concentrates on different techniques of predicting job satisfaction in different jobs.
Response 2: As a result of a better review of the state of the art, which we establish in section 2, now in the results chapter we discuss and make an analysis of our model with the models identified in section 2. We can verify this from line 343 up to line 391 of the pdf. Likewise, citations 21, 22 and 23 refer to studies related to the field of job satisfaction in teachers.
Point 3: The authors correctly state that "Since the data set is partially unbalanced this indicator could be misleading." However, they proceed to select some of these non-reliable indicators instead of using classification accuracy measures that are independent of the imbalance in the dataset (for example, Matthews correlation coefficient, Youden's J or Cohen's kappa). I suggest using indicators that are independent of the number of cases in the classes.
Response 3: Is right. In order to lift this observation, we have carried out an analysis based on indicators that are independent of the imbalance of our data set. This can be seen in figure 8, where we include the indicators suggested by the reviewer.
Point 4: The dataset is uploaded to Google Drive which is not permanent storage. I suggest publishing the dataset on the relevant sites and acquiring DOI for it.
Response 4: We changed it to the DOI:: https://doi.org/10.17632/b7wbthz6hs.2
Point 5: Figure 11 is difficult to read because many lines are very close to each other.
Response 5: Due to the corrections we have made, figure 11 is now figure 9. In it we can see the area under the ROC curve but separately according to the algorithm used.
Point 6: Variables, listed in Figure 2 and Figure 3, are not described in the article (there is no way to understand their designations from the article text).
Response 6: We have now added Appendix A.4 in which we can see the data dictionary of the features of our data set.
Point 7: I don't understand the reason for adding "Memory allocation" to Table 1 because it depends on the way the dataset is represented in the memory.
Response 7: You are right, we have deleted the indicated record.
Point 8: There are minor English problems (e.g., "It is currently known that satisfaction or dissatisfaction {of??} teachers with their work life...").
Response 8: We accept it. We have conducted a thorough review of the English language writing
Author Response File: Author Response.docx
Round 2
Reviewer 1 Report
I recommend one more time through to edit the English language. There are many long sentences that make the document difficult to read. Many times the document will read better if the semicolon is replaced with a period and a new sentence is created.
Line 250: "Choose" should be "chose"
Line 281: "Training" should be "train"
Nice work.
Author Response
Response to Reviewer 1 Comments
Point 1: I recommend one more time through to edit the English language. There are many long sentences that make the document difficult to read. Many times the document will read better if the semicolon is replaced with a period and a new sentence is created.
Response 1: .
Point 2: Line 250: "Choose" should be "chose"
Response 2: We appreciate the suggestion, we apologize for the mistake. We made the correction.
Point 3: Line 281: "Training" should be "train"
Response 3: We appreciate the suggestion, we apologize for the mistake. We made the correction.
Point 4: Nice work.
Response 4: Thanks a lot.
Author Response File: Author Response.docx
Reviewer 2 Report
All my comments have been attended to and effected.
The authors should delete empty reference 55 in line 587.
Author Response
Response to Reviewer 2 Comments
Point 1: All my comments have been attended to and effected
Response 1: We appreciate all your comments, which have allowed us to substantially improve the article.
Point 2: The authors should delete empty reference 55 in line 587.
Response 2: We did a review of the empty reference and fixed the bug.
Author Response File: Author Response.docx