pith. sign in

arxiv: 2605.24655 · v1 · pith:S4MKGALCnew · submitted 2026-05-23 · 📡 eess.SP

Simulation-Driven Ensemble Machine Learning for Robust and Generalizable Path Loss Prediction

Ahmed P. Mohamed , Byunghyun Lee , Yaguang Zhang , Christopher R. Anderson , David J. Love , James V. Krogmeier This is my paper

Pith reviewed 2026-06-30 12:56 UTC · model grok-4.3

classification 📡 eess.SP
keywords path loss predictionensemble machine learningsynthetic dataSMOTE augmentationlidar simulationwireless networkscross-environment generalizationdata scarcity
0
0 comments X

The pith

An ensemble model that blends lidar simulations with scarce real measurements and SMOTE cuts path loss prediction error by up to 50 percent while improving cross-environment generalization.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper establishes that an ensemble machine learning framework can overcome the scarcity of real path loss measurements by integrating them with synthetic data from a lidar-based simulator. The simulator supplies broad spatial coverage through static path loss values that reflect terrain and obstacles, while a dynamically weighted ensemble balances the two sources and SMOTE augments the real samples via interpolation. This combination, along with engineered propagation features, allows the model to capture geographical and physical variability across urban, suburban, residential, industrial, and rural settings. Experimental results show the approach yields up to 50 percent lower mean absolute error than real-data-only training and up to 25 percent better performance than synthetic-only training, especially when generalizing to new environments.

Core claim

The proposed simulation-driven ensemble framework integrates real measurements with synthetic path loss data from a lidar simulator, applies dynamic weighting to balance their contributions, incorporates SMOTE to synthesize additional samples from measurements, and uses engineered propagation features. This captures geographical and physical variability and enables adaptability across diverse environments, delivering up to a 50 percent reduction in mean absolute error relative to models trained solely on real data and up to 25 percent improvement relative to models trained exclusively on synthetic data, with the largest gains in cross-environment generalization.

What carries the argument

The dynamically weighted ensemble that integrates lidar-simulated synthetic path loss values with real measurements while balancing their relative contributions.

If this is right

  • The framework enables path loss prediction models to generalize across urban, suburban, residential, industrial, and rural environments despite limited measurements.
  • Wireless network planning and optimization become feasible with lower measurement costs by leveraging simulation coverage.
  • Data scarcity effects are mitigated through the combination of synthetic data, dynamic weighting, and SMOTE augmentation.
  • Engineered propagation features together with the ensemble allow the model to represent both geographical and physical variability in the propagation environment.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same simulation-plus-ensemble pattern could be tested on other wireless prediction tasks such as delay spread or interference levels where lidar or similar simulators already exist.
  • In deployment, the approach might shorten the measurement campaigns required before rolling out new base stations in unfamiliar regions.
  • Extending the simulator to include time-varying elements like moving vehicles could further improve performance in dynamic settings beyond the current static path loss focus.

Load-bearing premise

The lidar-based simulator supplies static path loss values that accurately capture terrain variations and physical obstacles across broad areas.

What would settle it

Running the trained ensemble on a new environment whose terrain and obstacle layout differ substantially from the simulator's training areas and observing no MAE reduction or outright worse performance compared with a real-data-only baseline.

read the original abstract

Machine learning has emerged as a promising approach to path loss prediction, yet its effectiveness often degrades when measurement data are scarce. To address this limitation, we propose an ensemble-based machine learning framework that integrates real measurements with synthetic data generated using a lidar-based simulator. The simulator provides broad spatial coverage through static path loss values that capture terrain variations and physical obstacles in the propagation environment. A dynamically weighted ensemble then combines simulation results with measured data, balancing the contribution of both data sources and improving generalization across diverse environments. To further mitigate the effects of limited measurements, we incorporate the Synthetic Minority Over-sampling Technique (SMOTE), a data augmentation technique that synthesizes additional samples through interpolation between measurements while preserving their statistical properties. By leveraging simulation data, SMOTE, and engineered propagation features, the proposed framework captures geographical and physical variability, enabling adaptability across urban, suburban, residential, industrial, and rural environments. Experimental results demonstrate that the proposed method achieves up to a 50% reduction in mean absolute error (MAE), compared with models trained solely on real data, and up to a 25% improvement relative to models trained exclusively on synthetic data, particularly for cross-environment generalization. These findings highlight the effectiveness of combining simulation-based synthetic data with SMOTE to overcome data scarcity and enhance the model's generalization ability. Overall, the proposed framework provides a robust and practical solution for path loss prediction across diverse environments with limited measurement data, supporting cost-effective planning and optimization of wireless networks.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The paper proposes an ensemble ML framework for path loss prediction that integrates scarce real measurements with synthetic data generated by a lidar-based simulator (providing static path loss values capturing terrain and obstacles), applies dynamic weighting to balance sources, uses SMOTE for augmentation, and incorporates engineered propagation features. It claims this yields up to 50% MAE reduction versus real-data-only training and 25% improvement versus synthetic-only training, with particular gains in cross-environment generalization across urban, suburban, residential, industrial, and rural settings.

Significance. If the performance claims hold under proper validation, the work would offer a practical approach to mitigating data scarcity in wireless propagation modeling by leveraging simulation for better generalization, which could support cost-effective network planning. The combination of lidar simulation, dynamic ensembling, and SMOTE is a coherent strategy for the stated problem, and the emphasis on cross-environment testing addresses a real limitation in the field.

major comments (2)
  1. [Abstract] Abstract: The central claims of up to 50% MAE reduction versus real-only models and 25% versus synthetic-only models (especially for cross-environment generalization) are presented without any details on dataset sizes, number of environments tested, baseline models, statistical significance tests, or error bars. This absence prevents assessment of whether the reported gains are robust or load-bearing for the paper's contribution.
  2. [Abstract] Abstract (and § on simulator description): The framework's ability to improve generalization rests on the assumption that the lidar-based simulator's static path loss values accurately encode propagation physics (terrain variations, physical obstacles) for unseen environments. No independent validation of these values against real measurements or standard models (e.g., in target environments) is described, raising the risk that reported gains derive primarily from increased data volume rather than transferable physical signal.
minor comments (1)
  1. [Abstract] The abstract mentions 'engineered propagation features' but does not list or define them explicitly; a table or subsection enumerating these features would improve reproducibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments, which highlight areas where additional clarity will strengthen the manuscript. We agree that the abstract requires more supporting details and that the simulator's fidelity merits explicit discussion. We will revise the abstract and add relevant content in the methods and results sections. Point-by-point responses follow.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The central claims of up to 50% MAE reduction versus real-only models and 25% versus synthetic-only models (especially for cross-environment generalization) are presented without any details on dataset sizes, number of environments tested, baseline models, statistical significance tests, or error bars. This absence prevents assessment of whether the reported gains are robust or load-bearing for the paper's contribution.

    Authors: We agree that the abstract as written does not supply the quantitative context needed to evaluate the claims. In the revised version we will expand the abstract to report the number of real measurements and synthetic samples used, the five environments tested (urban, suburban, residential, industrial, rural), the baseline models (real-only and synthetic-only ML regressors), and the statistical tests or error bars supporting the MAE reductions. These details already exist in the experimental section and will be summarized concisely in the abstract. revision: yes

  2. Referee: [Abstract] Abstract (and § on simulator description): The framework's ability to improve generalization rests on the assumption that the lidar-based simulator's static path loss values accurately encode propagation physics (terrain variations, physical obstacles) for unseen environments. No independent validation of these values against real measurements or standard models (e.g., in target environments) is described, raising the risk that reported gains derive primarily from increased data volume rather than transferable physical signal.

    Authors: The concern is valid. The current manuscript describes the lidar simulator's construction but does not present an independent validation (e.g., direct comparison of simulator outputs to held-out real measurements or to established models such as COST-231 in the target environments). To address this we will add a dedicated validation subsection that quantifies agreement between simulated and measured path loss where overlapping data exist, and we will discuss the implications for the observed generalization gains. This addition will clarify whether the performance improvement stems from physically grounded features or simply from larger data volume. revision: yes

Circularity Check

0 steps flagged

No circularity; claims rest on independent empirical comparisons

full rationale

The paper reports MAE reductions from training an ensemble ML model on combined real measurements, lidar-simulated synthetic data, and SMOTE-augmented samples, then evaluating on held-out cross-environment test sets. These are direct experimental outcomes, not quantities that reduce by construction to the simulator outputs or fitted parameters. No equations, self-citations, or uniqueness theorems are invoked to force the reported gains; the simulator is treated as an external data source whose fidelity is an assumption outside the derivation chain.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim depends on the simulator faithfully representing real propagation physics and on SMOTE producing statistically valid augmentations for path loss data; these are standard domain assumptions rather than new entities or free parameters beyond typical ML tuning.

free parameters (1)
  • Dynamic ensemble weights
    Weights balancing simulation and measurement contributions are learned or tuned dynamically during training.
axioms (1)
  • domain assumption Lidar-based simulator accurately models path loss variations due to terrain and obstacles
    Invoked when describing how the simulator supplies broad spatial coverage.

pith-pipeline@v0.9.1-grok · 5815 in / 1301 out tokens · 53550 ms · 2026-06-30T12:56:26.773637+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

66 extracted references · 6 canonical work pages

  1. [1]

    Simulation-enhanced data augmentation for machine learning pathloss prediction,

    A. P. Mohamedet al., “Simulation-enhanced data augmentation for machine learning pathloss prediction,” inICC 2024-IEEE Intl. Conf. Communications. IEEE, 2024, pp. 4863–4868

    2024

  2. [2]

    A survey of wireless path loss prediction and coverage mapping methods,

    C. Phillips, D. Sicker, and D. Grunwald, “A survey of wireless path loss prediction and coverage mapping methods,”IEEE Commun. Surveys Tuts., vol. 15, no. 1, pp. 255–270, 2012

    2012

  3. [3]

    Study on channel model for frequencies from 0.5 to 100 GHz,

    3GPP, “Study on channel model for frequencies from 0.5 to 100 GHz,” 3rd Generation Partnership Project (3GPP), Technical report (TR) 38.901, 01 2020, version 16.1.0

    2020

  4. [4]

    Two-dimensional deterministic propagation models approach and comparison with calibrated empirical models,

    M. Ayadi, N. Torjemen, and S. Tabbane, “Two-dimensional deterministic propagation models approach and comparison with calibrated empirical models,”IEEE Transactions on Wireless Communications, vol. 14, no. 10, pp. 5714–5722, 2015

    2015

  5. [5]

    A proposed mid-band statistical clutter prop- agation model utilizing lidar data,

    W. Kozma and M. Cotton, “A proposed mid-band statistical clutter prop- agation model utilizing lidar data,” in2023 17th European Conference on Antennas and Propagation (EuCAP). IEEE, 2023, pp. 1–5

    2023

  6. [6]

    Propagation modeling through foliage in a coniferous forest at 28 GHz,

    Y . Zhang, C. R. Anderson, N. Michelusi, D. J. Love, K. R. Baker, and J. V . Krogmeier, “Propagation modeling through foliage in a coniferous forest at 28 GHz,”IEEE Wireless Communications Letters, vol. 8, no. 3, pp. 901–904, 2019

    2019

  7. [7]

    Prediction of tropospheric radio trans- mission loss over irregular terrain: A computer method-1968,

    A. G. Longley and P. L. Rice, “Prediction of tropospheric radio trans- mission loss over irregular terrain: A computer method-1968,” U.S. De- partment of Commerce, Environmental Science Services Administration, Institute for Telecommunication Sciences, Tech. Rep. ERL 79-ITS-67, 1968, doi: 10.70220/d3k87mv1

    work page doi:10.70220/d3k87mv1 1968

  8. [8]

    Urban transmission loss models for mobile radio in the 900 and 1800 MHz band,

    COST 231, “Urban transmission loss models for mobile radio in the 900 and 1800 MHz band,” COST 231 TD (90), 119 Rev. 2, The Hague, The Netherlands, 1991

    1991

  9. [9]

    QuaDRiGa: A 3- D multi-cell channel model with time evolution for enabling virtual field trials,

    S. Jaeckel, L. Raschkowski, K. B ¨orner, and L. Thiele, “QuaDRiGa: A 3- D multi-cell channel model with time evolution for enabling virtual field trials,”IEEE Trans. Antennas Propag., vol. 62, no. 6, pp. 3242–3256, 2014

    2014

  10. [10]

    Deep learning- based signal strength prediction using geographical images and expert knowledge,

    J. Thrane, B. Sliwa, C. Wietfeld, and H. L. Christiansen, “Deep learning- based signal strength prediction using geographical images and expert knowledge,” inGLOBECOM 2020-2020 IEEE Global Commun. Conf. IEEE, 2020, pp. 1–6

    2020

  11. [11]

    An artificial in- telligence radio propagation model based on geographical information,

    H. Zhang, J. Dong, X. Liu, J. Liu, and X. Zhang, “An artificial in- telligence radio propagation model based on geographical information,” IEEE Trans. Antennas Propag., vol. 70, no. 12, pp. 12 049–12 060, 2022

    2022

  12. [12]

    Interpretable AI- based large-scale 3D pathloss prediction model for enabling emerging self-driving networks,

    U. Masood, H. Farooq, A. Imran, and A. Abu-Dayya, “Interpretable AI- based large-scale 3D pathloss prediction model for enabling emerging self-driving networks,”IEEE Trans. Mobile Comput., vol. 22, no. 7, pp. 3967–3984, 2022

    2022

  13. [13]

    Improving propagation model predic- tions via machine learning with engineered features,

    A. Vanleer and C. R. Anderson, “Improving propagation model predic- tions via machine learning with engineered features,” inMILCOM 2021- 2021 IEEE Military Communications Conference (MILCOM). IEEE, 2021, pp. 420–425

    2021

  14. [14]

    Challenges and opportunities of future rural wireless communications,

    Y . Zhang, D. J. Love, J. V . Krogmeier, C. R. Anderson, R. W. Heath, and D. R. Buckmaster, “Challenges and opportunities of future rural wireless communications,”IEEE Commun. Mag., vol. 59, no. 12, pp. 16–22, 2021

    2021

  15. [15]

    Analyzing and reducing the damage of dataset bias to face recognition with synthetic data,

    A. Kortylewski, B. Egger, A. Schneider, T. Gerig, A. Morel-Forster, and T. Vetter, “Analyzing and reducing the damage of dataset bias to face recognition with synthetic data,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops, 2019

    2019

  16. [16]

    Simulation-aided measurement-based channel modeling for propagation at 28 GHz in a coniferous forest,

    Y . Zhang, J. A. Tan, B. M. Dorbert, C. R. Anderson, and J. V . Krogmeier, “Simulation-aided measurement-based channel modeling for propagation at 28 GHz in a coniferous forest,” inGLOBECOM 2020- 2020 IEEE Global Commun. Conf.IEEE, 2020, pp. 1–6

    2020

  17. [17]

    Reciprocity-aware convolutional neural networks for map-based path loss prediction,

    R. G. Dempsey, J. Ethier, and H. Yanikomeroglu, “Reciprocity-aware convolutional neural networks for map-based path loss prediction,”arXiv preprint arXiv:2504.03625, 2025

    work page arXiv 2025

  18. [18]

    Mixing real and synthetic data to enhance neural network training–a review of current approaches,

    V . Seib, B. Lange, and S. Wirtz, “Mixing real and synthetic data to enhance neural network training–a review of current approaches,”arXiv preprint arXiv:2007.08781, 2020

    work page arXiv 2007

  19. [19]

    Multi-source domain adapta- tion: A causal view,

    K. Zhang, M. Gong, and B. Sch ¨olkopf, “Multi-source domain adapta- tion: A causal view,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 29, no. 1, 2015

    2015

  20. [20]

    Large- scale cellular coverage simulation and analyses for follow-me UA V data relay,

    Y . Zhang, J. V . Krogmeier, C. R. Anderson, and D. J. Love, “Large- scale cellular coverage simulation and analyses for follow-me UA V data relay,”IEEE Trans. Wireless Commun., 2023

    2023

  21. [21]

    T. G. Dietterich, Ensemble Methods in Machine Learning. In: Multiple Classifier Systems. MCS 2000. Lecture Notes in Computer Science, vol 1857. Springer, Berlin, Heidelberg, 2000, https://doi.org/10.1007/ 3-540-45014-9 1

    2000

  22. [22]

    Ensemble-based classifiers,

    L. Rokach, “Ensemble-based classifiers,”Artificial intelligence review, vol. 33, pp. 1–39, 2010

    2010

  23. [23]

    SMOTE: synthetic minority over-sampling technique,

    N. V . Chawla, K. W. Bowyer, L. O. Hall, and W. P. Kegelmeyer, “SMOTE: synthetic minority over-sampling technique,”J. Artificial Intelligence Research, vol. 16, pp. 321–357, 2002

    2002

  24. [24]

    3D spec- trum awareness for radio dynamic zones using Kriging and ma- trix completion,

    M. Hollingsworthet al., “Repurposing cellular reference signals: ac- curate RSRP measurements with mobile phones,” in2024 IEEE Intl. Symposium on Dynamic Spectrum Access Networks (DySPAN). IEEE, 2024, pp. 45–50. doi: 10.1109/DySPAN60163.2024.10632813

    work page doi:10.1109/dyspan60163.2024.10632813 2024

  25. [25]

    Spectrum Sharing Characterization Using Smartphones: Exploring 6 GHz Sharing Through Large-Scale Wi-Fi 6E Measurements,

    S. Do ˘gan-Tusha, A. Tusha, M. I. Rochman, H. Nasiri, and M. Ghosh, “Spectrum Sharing Characterization Using Smartphones: Exploring 6 GHz Sharing Through Large-Scale Wi-Fi 6E Measurements,”IEEE Communications Magazine, vol. 63, no. 2, pp. 70–76, 2025. doi: 10.1109/MCOM.001.2400325

    work page doi:10.1109/mcom.001.2400325 2025

  26. [26]

    G-NetTrack Pro,

    Gyokov Solutions, “G-NetTrack Pro,” https://www.gyokovsolutions. com/G-NetTrack%20Android.html, 2025

    2025

  27. [27]

    AT&T Mobility (United States of America) – Cellular Coverage and Tower Map,

    CellMapper.net, “AT&T Mobility (United States of America) – Cellular Coverage and Tower Map,” https://www.cellmapper.net/map?MCC=-1& MNC=1, 2025

    2025

  28. [28]

    Search for cell towers & antennas,

    AntennaSearch, “Search for cell towers & antennas,” https://www. antennasearch.com, 2025

    2025

  29. [29]

    Impact of 3D antenna radiation pattern in UA V air-to-ground path loss modeling and RSRP-based localization in rural area,

    S. J. Maeng, H. Kwon, O. Ozdemir, and G ¨uvenc ¸, Ismail, “Impact of 3D antenna radiation pattern in UA V air-to-ground path loss modeling and RSRP-based localization in rural area,”IEEE Open J. Antennas Propag., 2023

    2023

  30. [30]

    Measurement and Analysis of 4G/5G Mobile Signal Coverage in a Metropolitan Area,

    Y . Chenet al., “Measurement and Analysis of 4G/5G Mobile Signal Coverage in a Metropolitan Area,”Sensors, vol. 23, no. 15, p. 6594,

  31. [31]

    doi: 10.3390/s23156594

    work page doi:10.3390/s23156594

  32. [32]

    E-UTRA; Requirements for Support of Radio Resource Man- agement (Release 18),

    3GPP, “E-UTRA; Requirements for Support of Radio Resource Man- agement (Release 18),” 3rd Generation Partnership Project, Tech. Rep. TS 36.133, Jan. 2023, v18.0.0

    2023

  33. [33]

    Telecommunication management; LTE Radio Access Network (RAN) Performance Measurements,

    ——, “Telecommunication management; LTE Radio Access Network (RAN) Performance Measurements,” Tech. Rep. TS 28.658, 2021

    2021

  34. [34]

    E-UTRA; Further advancements for E-UTRA physical layer aspects (Release 9),

    ——, “E-UTRA; Further advancements for E-UTRA physical layer aspects (Release 9),” Technical Report TR 36.814, March 2010, v9.2.0

    2010

  35. [35]

    Sample-efficient learning of cellular antenna parameter settings,

    E. Tekgul, T. Novlan, S. Akoum, and J. G. Andrews, “Sample-efficient learning of cellular antenna parameter settings,” in2021 IEEE Inf. Theory Workshop (ITW). IEEE, 2021, pp. 1–6

    2021

  36. [36]

    Uplink-downlink joint antenna optimization in cellular systems with sample-efficient learning,

    ——, “Uplink-downlink joint antenna optimization in cellular systems with sample-efficient learning,” inIEEE Global Commun. Conf., 2022, pp. 6499–6504

    2022

  37. [37]

    Impact of electrical and mechanical antenna tilt on LTE downlink system performance,

    F. Athley and M. N. Johansson, “Impact of electrical and mechanical antenna tilt on LTE downlink system performance,” in2010 IEEE 71st Veh. Tech. Conf.IEEE, 2010, pp. 1–5

    2010

  38. [38]

    T-Mobile minor modification site application MC2021071498,

    Montgomery County Transmission Facilities Coordination Group (TFCG), “T-Mobile minor modification site application MC2021071498,” Montgomery County, Maryland, 2021. [On- line]. Available: https://montgomerycountytfcg.s3.amazonaws.com/ Applications/MC2021071498+Application.pdf

    2021

  39. [39]

    T-Mobile application: site filing for orange center road,

    Connecticut Siting Council, “T-Mobile application: site filing for orange center road,” 2020. [Online]. Available: https://portal.ct. gov/-/media/csc/2 ems-medialibrary/orange/orangecenterrd/t-mobile/ em-t-mobile-107-200817 filing 525-orange-ctr-rd-ct11412a.pdf

    2020

  40. [40]

    Beltsville Agricultural Research Center: replacement of three antennas at the T-mobile telecommunications facility,

    National Capital Planning Commission (NCPC), “Beltsville Agricultural Research Center: replacement of three antennas at the T-mobile telecommunications facility,” 2018. [Online]. Available: https://www.ncpc.gov/files/projects/2018/7979 Beltsville Agricultural Research Center Replacement of Three Antennas at the T-Mobile Telecommunications Facility - Water...

    2018

  41. [41]

    Verizon Wireless minor modification application: Site mc2021111599, Montgomery County, Maryland,

    Montgomery County Transmission Facilities Coordination Group (TFCG), “Verizon Wireless minor modification application: Site mc2021111599, Montgomery County, Maryland,” 2021. [Online]. Available: https://montgomerycountytfcg.s3.amazonaws.com/ Applications/MC2021111599+Application.pdf

    2021

  42. [42]

    Verizon Wireless colocation application: Site applica- tion MC2023102222, Montgomery County, Maryland,

    ——, “Verizon Wireless colocation application: Site applica- tion MC2023102222, Montgomery County, Maryland,” 2023. [Online]. Available: https://montgomerycountytfcg.s3.amazonaws.com/ Applications/MC2023102222+Application.pdf

    2023

  43. [43]

    JMA Wireless antenna solutions,

    JMA Wireless, “JMA Wireless antenna solutions,” 2025, accessed: January 2, 2025. [Online]. Available: https://www.jmawireless.com/

    2025

  44. [44]

    Antenna products and specifications,

    A. A. Solutions, “Antenna products and specifications,” 2025, accessed: January 2, 2025. [Online]. Available: https://amphenol-antennas.com/

    2025

  45. [45]

    CommScope antenna portfolio,

    CommScope, “CommScope antenna portfolio,” 2025, accessed: January 2, 2025. [Online]. Available: https://www.commscope.com/

    2025

  46. [46]

    CCI antenna solutions,

    Communication Components Inc. (CCI), “CCI antenna solutions,” 2025, accessed: January 2, 2025. [Online]. Available: https://www. cciproducts.com/

    2025

  47. [47]

    Ericsson antenna systems,

    Ericsson, “Ericsson antenna systems,” 2025, accessed: January 2, 2025. [Online]. Available: https://www.ericsson.com/

    2025

  48. [48]

    Report ITU-R SM.2028-2: Monte Carlo simulation methodology for the use in sharing and compatibility studies between different radio services or systems,

    International Telecommunication Union, “Report ITU-R SM.2028-2: Monte Carlo simulation methodology for the use in sharing and compatibility studies between different radio services or systems,” International Telecommunication Union, Geneva, Switzerland, Technical Report SM.2028-2, June 2017

    2028

  49. [49]

    Comparison of empirical propagation path loss models for fixed wireless access systems,

    V . S. Abhayawardhana, I. J. Wassell, D. Crosby, M. P. Sellars, and M. G. Brown, “Comparison of empirical propagation path loss models for fixed wireless access systems,” in2005 IEEE 61st Veh. Technol. Conf., vol. 1. IEEE, 2005, pp. 73–77

    2005

  50. [50]

    Study on 3D channel model for LTE,

    3GPP, “Study on 3D channel model for LTE,” Technical Report TR 36.873, June 2014, release 12

    2014

  51. [51]

    The extended Hata (eHata) urban propagation model,

    National Telecommunications and Information Administration, U.S. De- partment of Commerce, “The extended Hata (eHata) urban propagation model,” https://github.com/NTIA/ehata.git

  52. [52]

    [Online]

    Forsk,Atoll 3.3.2 User Manual for Radio Networks, France, 2016, aT332 UMR E0. [Online]. Available: https://github.com/ aprincemohamed/DeepLearningBasedCellularCoverageMap/raw/main/ Atoll-3-3-2-User-Manual-Radio.pdf

    2016

  53. [53]

    Indiana statewide normalized digital height model (2016-2019),

    J. Jung and S. Oh, “Indiana statewide normalized digital height model (2016-2019),” Feb 2021. [Online]. Available: https://purr.purdue.edu/ publications/3708/1

    2016

  54. [54]

    Geneva, Switzer- land: International Telecommunication Union, 2008, iTU-R Handbook, Edition 2008

    International Telecommunication Union,Radiowave Propagation Infor- mation for Designing Terrestrial Point-to-Point Links. Geneva, Switzer- land: International Telecommunication Union, 2008, iTU-R Handbook, Edition 2008

    2008

  55. [55]

    Machine learning-based mmWave path loss prediction for ur- ban/suburban macro sites,

    G. Reus-Muns, J. Du, D. Chizhik, R. Valenzuela, and K. R. Chowd- hury, “Machine learning-based mmWave path loss prediction for ur- ban/suburban macro sites,” inGLOBECOM 2022-2022 IEEE Global Communications Conference. IEEE, 2022, pp. 1429–1434

    2022

  56. [56]

    Propagation by diffraction,

    International Telecommunication Union, “Propagation by diffraction,” International Telecommunication Union, Recommendation ITU-R P.526- 16, Nov. 2025

    2025

  57. [57]

    Random forests,

    L. Breiman, “Random forests,”Machine learning, vol. 45, pp. 5–32, 2001

    2001

  58. [58]

    A decision-theoretic generalization of on-line learning and an application to boosting,

    Y . Freund and R. E. Schapire, “A decision-theoretic generalization of on-line learning and an application to boosting,”Journal of computer and system sciences, vol. 55, no. 1, pp. 119–139, 1997

    1997

  59. [59]

    Robust estimation of a location parameter,

    P. J. Huber, “Robust estimation of a location parameter,” inBreak- throughs in Statistics: Methodology and Distribution, S. Kotz and N. L. Johnson, Eds. New York, NY: Springer, 1992, pp. 492–518

    1992

  60. [60]

    Greedy function approximation: a gradient boosting machine,

    J. H. Friedman, “Greedy function approximation: a gradient boosting machine,”Annals of Statistics, pp. 1189–1232, 2001

    2001

  61. [61]

    XGBoost: A scalable tree boosting system,

    T. Chen and C. Guestrin, “XGBoost: A scalable tree boosting system,” in Proceedings of the 22nd ACM SIGKDD Intl. Conf. Knowledge Discovery and Data Mining, 2016, pp. 785–794

    2016

  62. [62]

    LightGBM: A highly efficient gradient boosting decision tree,

    G. Keet al., “LightGBM: A highly efficient gradient boosting decision tree,”Advances in Neural Inf. Process. Syst., vol. 30, 2017

    2017

  63. [63]

    CatBoost: unbiased boosting with categorical features,

    L. Prokhorenkova, G. Gusev, A. V orobev, A. V . Dorogush, and A. Gulin, “CatBoost: unbiased boosting with categorical features,”Advances in Neural Inf. Process. Syst., vol. 31, 2018

    2018

  64. [64]

    Goodfellow, Y

    I. Goodfellow, Y . Bengio, A. Courville, and Y . Bengio,Deep learning. MIT press Cambridge, 2016

    2016

  65. [65]

    Domain generalization in machine learning models for wireless communications: Concepts, state-of-the-art, and open issues,

    M. Akrout, A. Feriani, F. Bellili, A. Mezghani, and E. Hossain, “Domain generalization in machine learning models for wireless communications: Concepts, state-of-the-art, and open issues,”IEEE Communications Surveys & Tutorials, vol. 25, no. 4, pp. 3014–3037, 2023

    2023

  66. [66]

    Improving imbalanced learn- ing through a heuristic oversampling method based on K-means and SMOTE,

    G. Douzas, F. Bacao, and F. Last, “Improving imbalanced learn- ing through a heuristic oversampling method based on K-means and SMOTE,”Information sciences, vol. 465, pp. 1–20, 2018

    2018