Research on Student Performance Prediction Based on Stacking Fusion Model

Round 1

Reviewer 1 Report

- Article required more references, at least 30 lists should be cited. 

- The implication should be recommended. 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

In the article “Research on student performance prediction based on Stacking 2 fusion model”, the authors presented the Stacking fusion model based on RF-CART-XGBoost-LightGBM.

In the introduction, the authors reviewed the literature and, based on the research in the article, proposed a Stacking fusion model based on RF-CART-XGBoost-LightGBM. The first layer of the model selects the Decision Tree (CART), Random Forest, XGBoost and LightGBM four single models with better performance as base models. The second layer uses the LightGBM model. The authors select relevant data on online learning behavior as input to a model that aims to predict student outcomes. They then compare the Stacking fusion model with the four single models and the Bagging fusion model based on the four single models.

In the second chapter, the authors reviewed the following algorithms: CART decision tree, random forest, XGBoost, LightGBM.

In the third chapter, the authors presented the data that were used for the experiment. Five features were selected as input to the classification model. The authors also presented Bagging Performance Prediction Fusion Model, Stacking Performance Prediction Fusion Model.

Further on, the Authors presented an analysis of the results. From the experiments conducted, the proposed RF-CART-XGBoost-LightGBM Stacking fusion model has higher performance than the Bagging fusion model. Single models cause large errors and will result in error in data analysis compared to the Stacking Fusion model. Solves the accuracy and error rate of a single model. At the same time, both fusion models perform much better than single models such as Decision Tree (CART), Random Forest, XGBoost, and LightGBM. The RF-CART-XGBoost-LightGBM Stacking fusion model can effectively make up for the poor generalization capacity of the single algorithm model. The proposed stacking fusion model based on RF-CART-XGBoost-LightGBM improves the accuracy of performance prediction in educational data mining.

The authors planned further studies to analyze the effects of different methods of combining single models on the time and accuracy of fusion models in a more comprehensive dataset, and to determine the optimal method for combining single models.

 

The article covers the field and is correctly structured. The article is scientifically substantiated. The presented data allows us to positively evaluate the methods. However, I have some suggestions for the article:

- the literature review should also take into account the current studies in the field of RF-CART-XGBoost-LightGBM. It is also worth extending the literature review by more than 9 items;

- in the second chapter, it would be advisable to detail the selected algorithms (e.g. indication of advantages / disadvantages), which would make it possible to determine why these algorithms were selected by the Authors;

- the introduction or chapter 2 lacks the Bootstrap characteristics that were used in the experiment but not described;

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

Firstly, the author should carefully review the paper:
-There are some grammar errors and typos in the paper.
-Some figures are blurred and big figures may be given on a full page, for example, Figure 4.
Secondly, the author should give the details about the study:
-Introduction section should give the importance of the topic.
-The related works may be given in a separate section and the related work section should be comprehensive and should include more recent studies.
-The differences and novelties of the study should be given in detail.
The units and symbols in the equations must be given in related paragraphs.

Finally, I think the authors should address the questions above.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 3 Report

The paper now can be accepted for publication.