Horizontal Learning Approach to Discover Association Rules

Round 1

Reviewer 1 Report

This is an interesting study, in which a new approach is proposed to improve the performance of association rule learning. However, some issues need to be addressed or improved to improve the quality of the manuscript.

1) The introduction section is weak. Not enough background information (for instance, how association rule learning is applied in various engineering sectors) is presented to reveal the meaning of this study.

2) The proposed horizontal learning approach should be elaborated on systematically. Although the approach is applied to a small example, the working flows/procedures of the proposed approach are hard to capture in the manuscript. An overview of the working procedures is needed. For instance, a flowchart of the proposed algorithm may help.

3) The proposed approach was validated using hypothetical small examples. A valid and concrete case study is preferred. For instance, compare it to a case study from other literature or using a public dataset.

4) Figure formatting and language editing should be improved. For instance, axis labels are lacking in Fig. 2, and many tables lack row/column heads.

Language editing is necessary.

Author Response

Please, see the file

Author Response File: Author Response.docx

Reviewer 2 Report

In this paper, the authors addressed the issue and introduce a  method, which requires substantially lower run-time and memory requirements in comparison to  the methods presently in use (reduction from 𝑂(2 𝑚) to 𝑂 (2 𝑚 2 ) in the worst case). Generally, this is an interesting work. It can be accepted if the authors can consider the following issues: 1. Is the proposed approach suitable for other kinds of applications? 2. The quality of figures should be imporved. 3. Other kinds of intelligent algorithm are welcome to enrich the literature review such as WKN-OC: A New Deep Learning Method for Anomaly Detection in Intelligent Vehicles; 4. More examples are welcome to show the advantage of the proposed method.

In this paper, the authors addressed the issue and introduce a  method, which requires substantially lower run-time and memory requirements in comparison to  the methods presently in use (reduction from 𝑂(2 𝑚) to 𝑂 (2 𝑚 2 ) in the worst case). Generally, this is an interesting work. It can be accepted if the authors can consider the following issues: 1. Is the proposed approach suitable for other kinds of applications? 2. The quality of figures should be imporved. 3. Other kinds of intelligent algorithm are welcome to enrich the literature review such as WKN-OC: A New Deep Learning Method for Anomaly Detection in Intelligent Vehicles; 4. More examples are welcome to show the advantage of the proposed method.

Author Response

Please, see the file

Author Response File: Author Response.docx

Reviewer 3 Report

1. Paper summary

This paper designs a novel approach to discovering association rules through horizontal learning. The authors introduce this algorithm with some examples and compare the algorithm with existing methods in the related work section.

2. Strengths

• The paper aims to address an important problem - a novel algorithm to discover association rules.

• The writing of the paper is generally good and easy to follow.

3. Weaknesses

• The paper needs proper alignment, e.g., figure 3 are not in the center of the paper and Line 261 and 262 cannot be read.

• There are existing works [1] about horizontal partition to discover association rules and these may impact the novelty negatively.

• Some other algorithms mentioned in the related work may have a better time complexity performance while authors are not making comparisons in experiments.

4. Comments for authors

1. Significance

1. I think the algorithm design is correct which is able to promote the significance of the paper.

2. There can be an issue that the authors didn’t include a discussion of the space complexity and the accuracy of the algorithm.

2. Soundness

1. The major issue with the soundness of this paper is that it doesn’t compare the algorithm with other existing algorithms that may have better performance in time complexity through experiments.

3. Novelty

1. The main issue that can influence the novelty of this paper negatively is that there are horizontal partitions to discover association rules.

2. Another issue is that some other algorithms may have better time complexity such as the ECLAT algorithm.

4. Presentation

1. Authors may also want to discuss some use cases of the new algorithm and how it can benefit future research.

2. Authors may want to include some introductions of association rules and their usage in the first section.

3. Figure 1 doesn’t include the meaning of the X and Y axes.

5. Verifiability

1. The paper doesn’t discuss much to enhance its verifiability. The tool is also not published to be accessed by the public.

6. Some minor comments

1. All the figures in the paper are not vector graphics.

2. Line 235-237 should be 1, 2, 3 rather than 3, 4, 5

[1] Das, Anjan, Dhruba K. Bhattacharyya, and Jugal K. Kalita. "Horizontal vs. vertical partitioning in association rule mining: a comparison." Proceedings of the 6th International Conference on Computational Intelligence and Natural Computation (CINC). 2003.

Line 26: Analysing -> Analyzing

Line 80: therefore -> therefore,

Line 119: partitions -> partition's

Line 191: analyse -> analyze

Author Response

Please, see the file

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

All my comments have been addressed or answered. I have no further comments.

Reviewer 3 Report

Thank authors for responding to my comments and the new version of the paper addresses my comments.

The main concern now focuses on the evaluation of the experiments. In the paper, to evaluate an association rule, authors need to define a procedure by which they can measure the effectiveness/correctness of the rule and thus they define the coverage of the rule.

However, the simulation is not very sound because the paper doesn’t explain why they are using the newly designed metric, what is the difference between the new metric and the existing ones, and why authors prefer to not use the existing evaluations. I think the authors may want to add more evaluation experiments using the existing metrics such as support and confidence. Besides, there is no statistical significance test in the evaluation.

Another minor comment is that the abstract should also include a brief introduction to the algorithm.

Line 364: values -> value