Machine Learning-Based Paging Enhancement in 5G Network

Round 1

Reviewer 1 Report

In this manuscript, the authors present an alternative paging approach for 5G AMFs using machine learning and a Markov process for performance enhancement. They first present the KNN classification (compared to three ML methods) method in order to predict the group of UE subscribers and then they discuss the generation of state-transition probabilities in order to predict the gNodeB movement and perform the paging. Finally, they present results considering the signaling overhead and compare them to existing paging methods. The presented work is interesting and very comprehensive. A few minor comments:

·   The authors should discuss in the introduction general Machine Learning methods and their implementation in 5G networks and include some relevant references. For instance:

1.   Sevgican, Salih, et al. "Intelligent network data analytics function in 5G cellular networks using machine learning." Journal of Communications and Networks 22.3 (2020): 269-280.

2.   A. Giannopoulos et al., "Supporting Intelligence in Disaggregated Open Radio Access Networks: Architectural Principles, AI/ML Workflow, and Use Cases," in IEEE Access, vol. 10, pp. 39580-39595, 2022, doi: 10.1109/ACCESS.2022.3166160.

3.  Jiang, Chunxiao, et al. "Machine learning paradigms for next-generation wireless networks." IEEE Wireless Communications 24.2 (2016): 98-105.

·        You should also mention in the introduction additional methods for mobility prediction, e.g.:

4.  Jeong, Jaeseong, et al. "Mobility Prediction for 5G Core Networks." IEEE Communications Standards Magazine 5.1 (2021): 56-61.

·         Please keep the same text format (subsection 3.3.2, page 9 and page 10)

·         In my experience, Artificial Neural Networks (ANNs) usually outperform the other ML methods in terms of accuracy; however, they are typically more complex and require significant training time. Can you please comment on the above issue and possibly include this consideration in the manuscript?

Author Response

Please see the attachment

Author Response File: Author Response.docx

Reviewer 2 Report

The following problems need to be addressed to improve the quality of the paper.

(1) The reviewer wants to know how to apply the Markov process discussed in Section 3.2 to the proposed machine learning-based paging approach. For example, the states should be explicitly defined in the proposed machine learning-based paging approach.

(2) Section 3.3.1 claims “1000 UE profile data of every 10 min cycle observed for four days were applied to supervised learning”. However, Figure 5 shows the result of five days.

(3) The reviewer does not know how to realize the KNN, random forest, decision tree, and SVM in the proposed paging approach during the experiment. The parameter selection of the KNN, random forest, decision tree, and SVM is not described.

(4) Section 4.2 says “the probabilistic paging method applied with machine learning showed that the number of paging signals per second reduced by up to 25% owing to the probabilistic paging method involving a Markov process”. It is uneasy to confirm 25% reduction according to Figure 14.

(5) The writing quality of the paper should be improved. The grammar problems and typesetting errors need to be corrected. The examples include “The paging profile configuration shown in Table 1. has various paging-related settings for different UE service types” in Page 3, “Subscribers were classified into two UE group, as shown in Table 3” in Page 6, “The KNN do not need to any additional processing of the data, and the simple comparison with the reference provides a low-cost, simple but effective method” in Page 7, and “Table 5. Example of proposed AMF Paging Profile Configuration” in Page 10. It would be good if a fluent English-writer could go over the paper to correct errors.

Author Response

Please see the attachment

Author Response File: Author Response.docx

Round 2

Reviewer 2 Report

The new version does not address any problem proposed in the previous comments except adding two literatures in References. In my opinion, the quality of the paper is not good. As an example, we note the response “We think the Markov process is described with Figure 8 and Table 4 in 3.3.2 Proposed Process”. However, Figure 8 and Table 4 only give four transition probabilities. The Markov process at least should give a Markov matrix. In addition, in “The paging profile configuration shown in Table 1. has various paging-related settings for different UE service types” in Page 3, there is an extra period after “Table 1”. 

Author Response

Please see the attachment

Author Response File: Author Response.docx

Round 3

Reviewer 2 Report

English language still needs to be polished. I can point out some mistakes. “Herein, the effect of applying machine learning to the AMF paging process in a 5G core network is discussed” in Page 2 should be indented. “Each parameter of the four models has KNN as k = 20, random forest as entropy as gain and 100 as division measure, decision tree as entropy as gain and best as division, SVM as liner as kernel, Use 10 as normalization and 0.1 as gamma” in Page 7 is confusing, e.g., “SVM as liner as kernel”? “Use”?

Author Response

Thanks for your comments

I've updated my paper.