An Oversampling Method for Class Imbalance Problems on Large Datasets

Round 1

Reviewer 1 Report

I hope the following comments help authors improve their work:

1. The proposed method does not seem to work well for classes with a convex structure or minority classes that include multiple groups in the data space. Given that if the shape of the class is not spherical, the data generated for the minority class may be outside the class borders, or if the minority class includes two or more groups, the data generated will probably be very skewed. The authors are expected to address these challenges further. These cases have not been considered even in the examples provided in the manuscript.
2.  In the presented results, the amount of balance obtained by using different methods is not presented.
3. Instead of the AUC, it may be better to report the accuracy of minority class detection in order to have a better understanding of the quality of the data generated.

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 2 Report

The manuscript applsci-1616970 presents an oversampling method for large class imbalance problems. The manuscript is clear and well written with potential for publication in its actual form.

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 3 Report

This paper propose an oversampling method for class imbalance problems on large datasets, although the article does relevant experiments to demonstrate the effectiveness of the proposed algorithm, this paper still have some drawbacks:

1. This topic is about class imbalance problem, what's the research significance of this topic, and what‘s’ the application of this topic?
2. The target of this paper is to solve large datasets problem, but some experiments' datasets are not large datasets. The large datasets should be defined.
3. The comparison methods are relative old, please add more new comparison methods.
4. In 231 line shows the word mistake 'appl7ied', please check all the paper to avoid this kind of mistake.

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 4 Report

Thank you for your interesting paper.

There is a typo on the abstract, I think: “bluelarge datasets”.

The literature review is interesting and focused regarding the topic, which facilitates reading and stemming the contributions.

Also, the method is clearly described. There is plenty of details, including the key steps of the algorithm showing the differences to previous / standard approach.

The authors do not show the description of the abbreviation IR (I think it means imbalance ratio, but it needs to be clearly stated on the first reference).

However, I do have some concerns, because the authors state they validate on large datasets. Table 2 comprises small datasets, which is ok for a first experiment. Nevertheless, the datasets from table 5 seem more standard than large datasets, especially considering today’s world of Big Data. I would refrain from referring so often to large datasets.

Otherwise, the paper is interesting and, in my opinion, deserves to be published.

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Round 2

Reviewer 3 Report

I accept the revised version.