Deep Learning for Predicting Attrition Rate in Open and Distance Learning (ODL) Institutions
, Farida Shehu Garki 1, Jude Chukwuma Onyeakazi 3, Christiana Uchenna Ezeanya 1 and Elochukwu Ukwandu 4,*
Round 1
Reviewer 1 Report
Comments and Suggestions for Authorsplease find the attached file.
Comments for author File: 
Comments.pdf
Author Response
Comment 1: The topic is interesting of designing a deep-learning framework analyzing the student attrition rate. For my understanding, this is a binary classification problem. And both LSTM and lDCNN could be appropriate for this issue, while the presented in this study are somewhat ambiguous. Below are some suggestions for the authors' consideration.
The abstract needs to be revised because it uses a lot of words to describe student enrolment but disregards critical details on the problem of "student attrition" and its significance. And it should be highlight why it is a classification issue and the reason to use deep leaning approaches.
Response 1: Thanks for this comment. We appreciate it all. The Abstract has been revised to help improve the quality of the manuscript as pointed out in your comment. For instance, in lines 16, 17 and 20.
Comment 2: The dataset should be detailed explained to prove the techniques and analyses used in the study are appropriate.
Response 2: Thanks for this comment. There is an improvement on the explanation of dataset as highlighted in the Methodology (lines 114-118).
Comment 3: The figures in the manuscript are displayed in an unprofessional manner; please resize them to a higher resolution and maintain word font consistency.
Response 3: Thanks for this comment. The figures have been improved on in terms of resizing and quality.
Comment 4: Key details for the loss/accuracy plots from Figures 12 to Figure 17 are absent for the x and y axis definitions. Additionally, as we do not observe an exponential drop or exponential increase for the training process, the training findings for LSTM and lDCNN are not convincing. Please verify the algorithms' hyperparameters. The data shown in Figures 12 through 17 only indicate that training has not reached convergence.
Response 4: Thanks for this comment. These has been improved on to help improve the quality of the manuscript as pointed out.
Comment 5: Training costs should be given for the two models.
Response 5: Thanks for this comment, but we want to admit that we are not really sure about the cost being sought out for here – financial, resources or time. We used the available free resource within reach to carry out the research.
Comment 6: The reason LSTM is a better fit for this problem than lDCNN needs additional in-depth examination. As far as I'm aware, lDCNN shouldn't perform this poorly for this particular issue.
Response 6: Thanks for this comment. We have reported as the result we received within the ambit of the meagre resources within our reach. We recognise there are certain constraints/impacts that our meagre resources may pose to the outcome of this research, but we want to believe that it is very minimal to the best of our knowledge.
Comment 7: The manuscript reads like a first draft, rather than something that has undergone some degree of f polishing and internal review before being sent out to the journal.
Response 7: Thanks for this comment. The authors have collective helped in revising the manuscript to help improve on it.
Reviewer 2 Report
Comments and Suggestions for AuthorsThe authors used LSTM model to investigate factors contributing to attrition and compared its performance with 1DCNN models, which indicates a superior correct classification rate compared to the 1DCNN model. The author has done a great job, but there are several issues to be aware of:
1. The author should change some pictures and tables to make them clearer and more standardized (For example, the text of "Figure 1" is fuzzy, "6.5%" of "Figure 2" is not marked, the text above "Figure 3" is too small, the picture of "Figure 6" is mirrored, and the text description of "Figure 9" is "illness" while the notes are "sickness", The diagram for "Figure 12" is too smalll, and" Table 4 "lacks the content of CNN);
2. The author should indicate the classification basis for Table1 and Table2 in the paragraph below line 168, and add illustrations to the description of the model in sections 3.4.1 and 3.4.2 for a clearer explanation;
3. The author should check for and correct the number of figure in line 234 and add a corresponding illustration to the paragraph in line 242;
4. The author should state "institutianal challenge" and "non-institutional challenge" in the text description of the paragraph in line 312 and the paragraph in line 356, respectively, otherwise too much of the same description may lead to misunderstanding;
5. The author should divide the conclusion paragraph into sections or add subheadings to enhance the sense of hierarchy of the conclusion content.
6. Some related works are missing.[1] Deep-IRTarget: An automatic target detector in infrared imagery using dual-domain feature extraction and allocation [2] Cognition-Driven Structural Prior for Instance-Dependent Label Transition Matrix Estimation
Comments on the Quality of English LanguageNAN
Author Response
Comment 1: The authors used LSTM model to investigate factors contributing to attrition and compared its performance with 1DCNN models, which indicates a superior correct classification rate compared to the 1DCNN model. The author has done a great job, but there are several issues to be aware of:
The author should change some pictures and tables to make them clearer and more standardized (For example, the text of "Figure 1" is fuzzy, "6.5%" of "Figure 2" is not marked, the text above "Figure 3" is too small, the picture of "Figure 6" is mirrored, and the text description of "Figure 9" is "illness" while the notes are "sickness", The diagram for "Figure 12" is too smalll, and" Table 4 "lacks the content of CNN);
Response 1: Thanks for pointing these out. In line with the above, we have improved on the fonts and picture quality as highlighted although the manuscript.
Comment 2: The author should indicate the classification basis for Table1 and Table2 in the paragraph below line 168, and add illustrations to the description of the model in sections 3.4.1 and 3.4.2 for a clearer explanation.
Response 2: We appreciate this comments. Please do see highlights on Sections 3.2 and 3.3 for our efforts in revising this manuscript in line with your comments.
Comment 3: The author should check for and correct the number of figure in line 234 and add a corresponding illustration to the paragraph in line 242.
Response 3: Thanks for this comment, We want to trust that our efforts on those line meets your expectation with regards to the issues raised.
Comment 4: The author should state "institutianal challenge" and "non-institutional challenge" in the text description of the paragraph in line 312 and the paragraph in line 356, respectively, otherwise too much of the same description may lead to misunderstanding.
Response 4: Thanks for these, the challenges have been made obvious in Sections 3.2 and 3.3 as highlighted.
Comment 5: The author should divide the conclusion paragraph into sections or add subheadings to enhance the sense of hierarchy of the conclusion content.
Response 5: Thanks for these comments. The division has been affected as highlighted by separating Conclusion, Limitations of Study and Recommendations.
Comment 6: Some related works are missing.[1] Deep-IRTarget: An automatic target detector in infrared imagery using dual-domain feature extraction and allocation [2] Cognition-Driven Structural Prior for Instance-Dependent Label Transition Matrix Estimation
Response 6: Thanks for this comment as we appreciate it all. Please see Table 1 as highlighted with improvements on the related works.
Reviewer 3 Report
Comments and Suggestions for AuthorsSpecifically, the enhancements in the introduction and the thorough explanation of the methodology are commendable. Additionally, the inclusion of a more detailed discussion on the limitations of the study provides a more balanced perspective.
Specific Comments:
Introduction:
- The introduction effectively sets the context for the study. However, it would benefit from a more detailed discussion on the current state-of-the-art in attrition prediction using machine learning. This would help to position your work more clearly within the existing body of research.
Results and Discussion:
- The results are presented clearly with appropriate use of tables and figures. However, the discussion could be enriched by comparing your findings with those of similar studies. How do your results compare with previous research? This would provide a broader context for your findings.
- The study mentions the superior performance of the LSTM model compared to the 1DCNN model. It would be helpful to provide a more detailed analysis of why this might be the case, perhaps discussing the strengths and weaknesses of each model in the context of the specific dataset used.
Minor Comments:
- Ensure consistency in terminology throughout the manuscript. For instance, "student dropout" and "student attrition" should be used consistently.
- Proofread the manuscript to correct minor grammatical errors and improve readability.
Author Response
Comment 1: Specifically, the enhancements in the introduction and the thorough explanation of the methodology are commendable. Additionally, the inclusion of a more detailed discussion on the limitations of the study provides a more balanced perspective.
Response 1: Thanks for your kind words. We deeply appreciate it all.
Comment 2: The introduction effectively sets the context for the study. However, it would benefit from a more detailed discussion on the current state-of-the-art in attrition prediction using machine learning. This would help to position your work more clearly within the existing body of research.
Response 2: We appreciate your pointing this out. In line with your comment, lines 31-35 as highlighted has been added to help address your concern. Trust that this meets the expected revision required.
Comments 3: The results are presented clearly with appropriate use of tables and figures. However, the discussion could be enriched by comparing your findings with those of similar studies. How do your results compare with previous research? This would provide a broader context for your findings.
Response 3: Thanks for these comments, it has helped in enriching our work. Please highlights in lines 406-407
Comment 4: The study mentions the superior performance of the LSTM model compared to the 1DCNN model. It would be helpful to provide a more detailed analysis of why this might be the case, perhaps discussing the strengths and weaknesses of each model in the context of the specific dataset used.
Response 4: We appreciate this comments. We have made some revision as highlighted in subsections 3.4.1 and 3.4.2 as highlighted.
Comment 5: Ensure consistency in terminology throughout the manuscript. For instance, "student dropout" and "student attrition" should be used consistently.
Response 5: The issue of consistency in writing has been taken care of in the manuscript to make for ease of comprehension.
Comment 6: Proofread the manuscript to correct minor grammatical errors and improve readability.
Response 6: Thanks for your comment. A proofread has been done to improve the quality of the manuscript.
Round 2
Reviewer 2 Report
Comments and Suggestions for AuthorsThe author addressed most of my concerns, except Comment 6.
The author addressed most of my concerns, except Comment 6.
Author Response
Comment 6: Proofread the manuscript to correct minor grammatical errors and improve readability.
Response 6: We are so sorry that there are some other minor omissions and grammatical errors still found in the manuscript. We have now patiently and thoroughly proofread the manuscript and have made amends to some of the earlier omitted errors including Figure numbering error that was earlier omitted in the manuscript. We really appreciate the reviewer for bearing with us by pointing out these omissions again to us in this round. The authors want to believe that the readability of the manuscript has been greatly improved by this effort in round 2 of the review process.
