Dichotomization of Multilevel Variables to Detect Hidden Associations
, Lisbeth Rodriguez-Mazahua 2, Farid GarcÃa-Lamont 3, Maricela Quintana-López 4 and Carlos A. Rojas-Hernández 1
Round 1
Reviewer 1 Report
The authors in this paper propose a new method to provide a test of independence to statistical data analysis for models that have multilevel variables. The paper is well written and organised, and touches on an important and very interesting topic.
It is suggested the authors develop a dedicated related work section byound the introduction section. The conclusions should also include how this approach can be extended, what advantages/disadvantages their algorithm has, etc.
Author Response
We appreciate your comments on the document.
We have carried out a review of the literature in search of a similar method; however, we have only found two papers (by the authors Sharpe and Zheng) that use an approach like the one proposed. Both articles have been cited in the introduction of the new version of our article.
Now, we have included the advantages and disadvantages of the method in the conclusions. Line 278.
Reviewer 2 Report
1. Since the most obvious application of the proposed association detection approach lies in the field of machine learning, I recommend reformatting the article from a machine learning perspective.
2. It is necessary to formulate the dependence of the change in calculation time on the number of variables.
3. For the considered databases, it is advisable to build correlation matrices and compare them with the results of determining associations.
Author Response
Thank you very much for your observation. Although the chi-square test is used in machine learning (mainly for attribute selection) the implementation shown in the article is intended to serve as a tool to discover associations between multilevel categorical variables. We hope that the method and the proposal will be used in Applied Sciences of any area, not only in machine learning.
A complexity analysis has been now included in the paper. Line 167.
It was not possible to calculate a correlation matrix with the data sets, since the variables are of the categorical type.
Reviewer 3 Report
The paper makes an interesting contribution to the literature, and well-organized and easy to follow. So, I believe that it is qualifies for publication.
Author Response
We highly appreciate your observation.
