A Survey of Machine Learning Approaches and Techniques for Student Dropout Prediction

Tracking #: 539-1519

Authors:

	Name	ORCID
	Neema Mduma	https://orcid.org/0000-0002-4364-3124
	Khamisi Kalegele	https://orcid.org/0000-0003-3020-6408
	Dina Machuve	https://orcid.org/0000-0002-8711-5948

Responsible editor:

Richard Mann

Submission Type:

Survey Paper

Abstract:

School dropout is absenteeism from school for no good reason for a continuous number of days. Addressing this challenge requires a thorough understanding of the underlying issues and effective planning for interventions. Over the years machine learning has gained much attention in addressing the problem of students dropout. This is because machine learning techniques can effectively facilitate determination of at-risk students and timely planning for interventions. To this end, several machine learning algorithms have been proposed in literature. This paper presents a survey of machine learning in education with focus on approaches and techniques for student-dropout prediction. Furthermore, the paper discusses the state of student dropout in developing countries and several performance metrics used by researchers to evaluate machine learning techniques in the context of education with experimental case study. Finally, the paper highlights challenges and future research directions.

Manuscript:

ds-paper-539.pdf

Previous Version:

A survey of Machine Learning Approaches and Techniques for Student Dropout Prediction

Data repository URLs:

http://www.twaweza.org/go/uwezo-datasets

Date of Submission:

Monday, July 23, 2018

Date of Decision:

Monday, September 10, 2018

Nanopublication URLs:

Decision:

Reject

Solicited Reviews:

Review #1 submitted on 28/Jul/2018

Review Details

Reviewer has chosen to be Anonymous

Overall Impression: Average
Suggested Decision: Undecided
Technical Quality of the paper: Weak
Presentation: Average
Reviewer`s confidence: High
Significance: Moderate significance
Background: Reasonable
Novelty: Limited novelty
Data availability: With exceptions that are admissible according to the data availability guidelines, all used and produced data are FAIR and openly available in established data repositories
Length of the manuscript: The length of this manuscript is about right

Summary of paper in a few sentences (summary of changes and improvements for second round reviews):

The paper is entitled as a survey but there is also a case study in the manuscript.
I think that the authors should rewrite the manuscript and mention as main contribution the case study.

Reasons to accept:

There is a case study.

Reasons to reject:

There are missing very well know citations as a survey paper
The authors should also examining other well known techniques for handling imbalanced datasets.

Nanopublication comments:

Further comments:

Review #2 submitted on 05/Sep/2018

By Lavi Aulck ORCID logo

https://orcid.org/0000-0001-6824-3875

Review Details

Reviewer has chosen not to be Anonymous

Overall Impression: Weak
Suggested Decision: Reject
Technical Quality of the paper: Weak
Presentation: Weak
Reviewer`s confidence: Medium
Significance: Moderate significance
Background: Reasonable
Novelty: Limited novelty
Data availability: All used and produced data (if any) are FAIR and openly available in established data repositories
Length of the manuscript: The length of this manuscript is about right

Summary of paper in a few sentences (summary of changes and improvements for second round reviews):

The paper examines the degree to which various ML approaches can be used to predict student dropout. It does so by running a suite of ML algorithms to predict dropout using a publicly-available dataset of tens of thousands of samples from Tanzania.

Reasons to accept:

The paper addresses an important area, particularly when using data from more traditional (i.e. non-online) sources. The authors' focus on developing countries also adds to the uniqueness of the work.

Reasons to reject:

The grammar has been improved from the previous version, though there are still some instances where editing is needed. The use of cross validation is not well detailed - why the need for separate validation and test sets with cross validation? Additionally, most of the results are presented in the figures, but they aren't described or discussed in detail in the text. Finally, figure 13 is missing, I believe. The different over-/under-sampling techniques should also be detailed. I would also like to see more detail with respect to highly important features in the ML models and any association these would have to practice/theory.

Nanopublication comments:

Further comments:

RESPONSE TO REVIEWERS

Dear Editor,

Re: Rebuttal Letter for Manuscript 535-1515 "A Survey of Machine Learning Approaches and Techniques for Student Dropout Prediction"

Reference is made to the above mentioned manuscript submitted to your journal and valuable reviewers comments that we received.
All the comments have been addressed and detailed on this rebuttal letter.
The corresponding modified manuscript in response to the reviewers’ comments is attached.

The reviewers’ comments have been addressed as follows:

1. The writing for this work is in need of proofing. There are numerous grammatical errors as well as latex formatting errors. This often stands as an obstacle in understanding what the writers are presenting.

Response:
The grammatical are corrected and latex formatting errors are addressed on the whole manuscript in line with the reviewer’s comments.

2. The introduction section discusses numerous topics that are not discussed in greater detail later in the writing (namely, gender disparities with respect to dropout)

Response:
Gender is discussed in subsections 4.6 and 5.1 of the paper, the feature importance experiment indicated the dropout rate association with gender. Student gender is considered a feature on the experiment, indicated on Table 3.

3. The paper initially presents itself as a review/survey but does not present much more than a typical literature review. Furthermore, the authors seems caught between attempting to present a standalone review/survey of prior work and presenting their own machine learning experiment(s). I'd suggest the authors stick to one or the other.

Response:
We agree with reviewer’s comment and present the manuscript as survey, we have rephrased all sections and describe the experimental part as case study.

4. An additional discussion of the methods in the ML experiments is needed. What does the dataset look like? Which features are present? Why were those specific ML algorithms selected for use? What was the train/test split?

Response:
In response to this comment we reconstruct the experimental framework section 4 and in line with reviewer’s comment. Furthermore, we present Table 3 that clearly describe features presented with additional data type details.

5. The authors only glance over their findings/results from both their review and their experiments. I think an additional discussion of either/both is needed.

Response:
We agree with reviewer’s comment and provide additional discussion on experiments in section 4.

6. The authors should make a better survey of Machine Learning Approaches for Student Dropout Prediction. e.g. Carlos Márquez-Vera, Alberto Cano, Cristóbal Romero, Amin Y. Noaman, Habib M. Fardoun, Sebastián Ventura: Early dropout prediction using data mining: a case study with high school students. Expert Systems 33(1): 107-124 (2016) as well as about the procedure for handling imbalance datasets e.g Krawczyk, B. (2016). Learning from imbalanced data: open challenges and future directions. Progress in Artificial Intelligence, 5(4), 221-232.

Response:
We surveyed these two publications and incorporated in subsections 4.3 and 5.1 of our manuscript.

7. The experimental framework is not well designed. The results are based on only one dataset. Overfitting should be included.

Response:
We agree with reviewer that the study based only on one dataset, we elaborate the experimental procedure in subsection 4.2 and present the experiment as case study.

8. Section 3.3 is not really useful.
The results are based on only one dataset.
If the authors want to propose an algorithm should present it with pseudocode.
A statistical test should be used for the comparison of the examined methods.
Some information about the time efficiency of the proposed method should be included.

Response:
We in line with reviewer’s comment and included additional details in subsection 5.1. The survey doesn’t propose any algorithm but evaluating the best algorithms based on the conducted experiment. Therefore, we didn’t include pseudocode for any algorithm presented. The additional details provide more clarification on experimental procedure.

We look forward to hearing from you regarding our submission. We would be glad to respond to any further questions and comments that you may have.

2 Comments

Looks Better

Submitted by Colleen Farrelly on Mon, 07/23/2018 - 16:41

Authors seem to have addressed some of the concerns. However, the paper still seems to be diverging into two topics--a review of machine learning methods and a review of educational data analysis. Doesn't work well as both, with text short-changing both topics to fit into a single paper. The machine learning algorithms being considered aren't described in detail, and the rationale for choosing them isn't clear. Should either be an education paper or revised as a machine learning paper.

Meta-Review by Editor

Submitted by Tobias Kuhn on Mon, 09/10/2018 - 01:12

Your manuscript was re-reviewed by two expert referees. Their reviews indicate that they are not convinced that the weaknesses identified in the first round of review have been adequately addressed. In particular, it is still ambiguous whether the manuscript is aiming to survey the field or to present a particular case study. In the case of the former, the reviewers indicated that a more detailed review of key material is required. In the case of the latter, the reviewers indicated that the methodology for the case study needs to be described with greater precision and full justification of all methods.

Richard Mann (http://orcid.org/0000-0003-0701-1274)

Data Science