Neural-Guided Adaptive Clustering for UAV-Based User Grouping in 5G/6G Post-Disaster Networks

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This paper introduces a hybrid clustering framework that dynamically selects between APC and density-based clustering (DBSCAN), guided by a neural classifier trained on spatial distribution features. The chosen centroids then seed a Genetic Algorithm that evolves UAV trajectories under multiple performance indicators, including coverage, capacity, and path efficiency.

My comments are as follows:

 

1. The motivation of this paper can be further highlighted to identify the main problem in UAV-Based User Grouping.
2. As a research paper, the authors describe too much in related work. It should be reduced and simplified.
3. In page 15, noise was set to -112 dBm. Any theoretical support for this value?
4. In page 18, “The system dynamically adjusts these parameters in response to environmental and operational factors, including weather conditions such as wind speed, visibility, and temperature” How to guarantee the parameters can reasonably reflect the situation of real environment.
5. Some latest progresses, such as, smart collaborative evolvement for virtual group creation in customized industrial IoT, enhanced user clustering and pairing scheme for NOMA-aided UAV networks, should be added and analyzed.
6. More analysis should be provided for Figure 4. Intersection Ratio (Line Crossing Avoidance).

Author Response

Please refer to attached document for our responses to Reviewer 1

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

The approach of a hybrid clustering architecture guided by machine learning to optimize user grouping and UAV routing is relevant, innovative, and well aligned with current challenges. The topic is timely and significant, given the increasing importance of UAVs and 5G/6G technologies in post-disaster recovery. The combination of adaptive clustering with genetic optimization and neural networks is novel and appears suitable for dynamic and heterogeneous environments. The authors are commended for the substantial amount of relevant information summarized in the tables. However, some weaknesses and concerns affecting the clarity, generalizability, and validation of the work can be identified.

1. The motivation for specifically choosing APC and DBSCAN as the two base methods is not entirely clear, nor is the rationale for excluding other modern or more robust clustering techniques. The connection between the contributions and prior work should be strengthened with explicit references.

2. The justification for the hybrid architecture and the need for the neural selector requires further depth. Why does the rigid approach fail, and how does the neural network improve upon this beyond the hypothesis of better adaptation? A detailed description of the hybrid scheme is suggested, possibly through block diagrams or algorithmic representations, clarifying the objectives pursued at each step.

3. The system is validated through simulations, but it is recommended to address or at least discuss in detail the UE distributions used and realistic scenarios that adequately represent variants such as evacuations or physical obstacles, to enrich practical applicability.

4. The impact of risk factors such as interference, latency, and rapid temporal variability of users is not discussed.

5. The results and conclusions tend to generalize the benefits broadly without recognizing specific limitations or scenarios in which the method might perform suboptimally.

6. The mathematical rigor should be improved concerning the explanation of equations, definitions of all variables, and appropriate punctuation in equations (using "," or "." as applicable).

7. The figures should be better supported by their respective captions. Figure 2, in particular, requires a more effective explanation, as it is currently difficult to interpret.

Author Response

Please refer to attached document for our responses to Reviewer 2

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

The authors of this study present a neural-guided hybrid clustering framework for UAV-assisted 
communication recovery in post-disaster failed networks. The study is well written, but some minor corrections should be made. The authors must define the parameters, physical quantities, acronyms (and so on) when they are first met in the text. For example, PSO is defined after the third mention. Fit-FCM wasn't defined at all (Fitness-based Fuzzy C-Means). In Eq. 1, you also didn't define Bw at all. Anyway, I would think that "B" as a notation for bandwidth is more appropriate. Although the Okumura-Hata model is well-known, it is customary to define all the parameters: f - frequency, hb - transmitter antenna height, and so on. Moreover, I feel that some inaccuracies are present. Such is the case at line 374, where the sentence is not very clear: 200 kbps needed for every UE is the bandwidth or the throughput? 

Author Response

Please refer to attached document for our responses to Reviewer 3

Author Response File: Author Response.pdf

Reviewer 4 Report

Comments and Suggestions for Authors

The paper deals with a method for improving UAVs clustering and trajectory path in an emergency scenario in case of failure of multiple base stations. The paper has an excellent review of the related literature.

Main comments:

• The selection of the clustering method and the clustering itself is assisted by neural networks algorithms. The training and the simulations are performed on a given system model, with a number of UEs, a given area and other parameters. In these cases, a general issue is the sensitivity of the system performance w.r.t. its application in different settings conditions. Therefore, more details should be given about the training of the ANNs and then results on scenarios with different parameters should be given and commented. 
•  The channel model is also a critical issue in the system model. It is used the Okumura-Hata model, but this is validated for ground communications and it is not clear if it used also for ground-to-UAV links. Ground-to-UAV Communication should be modelled by alternative models.
• The comparison with a standard benchmark is made through the affinity propagation clustering (APC) could be completed with the mentioned and well-known K-means.  

Author Response

Please refer to attached document for our responses to Reviewer 4

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

All my suggestions and concerns were satisfactorily addressed.

Author Response

Thank you for your comments, we have improve the write up of Section 3.2, 4.5, and 4.4.4.

Reviewer 4 Report

Comments and Suggestions for Authors

The authors have answered to the issues raised during the first review round.

There are still some point to clarify or address:

• In Sect. 4.4 it is stated that "cross-scenario validation where the ANN trained on one setup was tested on different scales, UE densities, and BS distributions". However this part seems not present in the text.
• Sect. 4.5 is missing.
• The response about the channel model "... the framework is agnostic to the propagation model. A sensitivity check demonstrated consistent trends across both channel models" should be addressed and commented also in the text.

Author Response

Please see the attached file 

Author Response File: Author Response.pdf