Evaluating Single- and Multi-Topology XGBoost Models for CSRR Inverse Design

Leonardo O. De Sousa; Mirelli S. Da Silva; Gabriel G. Machado; Francisco De A. Brito Filho; Samanta M. De Holanda; Isaac B. T. Da Silva

doi:10.1109/lacap67012.2026.11471168

Evaluating Single- and Multi-Topology XGBoost Models for CSRR Inverse Design

Leonardo O. De Sousa
, Mirelli S. Da Silva
, Gabriel G. Machado
, Francisco De A. Brito Filho
, Samanta M. De Holanda
, Isaac B. T. Da Silva

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

Abstract

This work investigates XGBoost-based inverse-design models for circular and square CSRRs. Single-topology models achieved the best performance, with MAEHFSS of 0.0457 and 0.0443 GHz for the circular and square cases, respectively, and low standard deviations across bandwidth variations. In contrast, multi-topology models showed large deviations, with MARHFSS reaching 0.3686 GHz and sensitivity S=1.44. The average S21 responses further confirmed that model C produced resonances closest to the 2.45 GHz target. These results demonstrate that topology-specific training provides more accurate and stable CSRR designs.

Original language	English
Title of host publication	2026 IEEE 2nd Latin American Conference on Antennas and Propagation (LACAP)
Publisher	IEEE
Pages	1-2
Number of pages	2
ISBN (Electronic)	979-8-3315-9779-5
ISBN (Print)	979-8-3315-9780-1
DOIs	https://doi.org/10.1109/lacap67012.2026.11471168
Publication status	Published online - 13 Apr 2026
Event	2026 IEEE 2nd Latin American Conference on Antennas and Propagation - Natal, Brazil Duration: 22 Feb 2026 → 25 Feb 2026

Conference

Conference	2026 IEEE 2nd Latin American Conference on Antennas and Propagation
Abbreviated title	LACAP
Country/Territory	Brazil
City	Natal
Period	22/02/26 → 25/02/26

Keywords

Antennas and propagation
Microwave antennas
Equivalent circuits
Filtering
Circuits
Filters
Protocols
HTTP
Wireless Access in Vehicular Environments
Machine learning
Extreme Gradient Boosting
XGBoost
Complementary Split Ring Resonator
CSRR

Access to Document

10.1109/lacap67012.2026.11471168

Cite this

De Sousa, L. O., Da Silva, M. S., Machado, G. G., De A. Brito Filho, F., De Holanda, S. M., & Da Silva, I. B. T. (2026). Evaluating Single- and Multi-Topology XGBoost Models for CSRR Inverse Design. In 2026 IEEE 2nd Latin American Conference on Antennas and Propagation (LACAP) (pp. 1-2). IEEE. Advance online publication. https://doi.org/10.1109/lacap67012.2026.11471168

@inproceedings{043f1720e94945d28f6a014d066a0058,

title = "Evaluating Single- and Multi-Topology XGBoost Models for CSRR Inverse Design",

abstract = "This work investigates XGBoost-based inverse-design models for circular and square CSRRs. Single-topology models achieved the best performance, with MAEHFSS of 0.0457 and 0.0443 GHz for the circular and square cases, respectively, and low standard deviations across bandwidth variations. In contrast, multi-topology models showed large deviations, with MARHFSS reaching 0.3686 GHz and sensitivity S=1.44. The average S21 responses further confirmed that model C produced resonances closest to the 2.45 GHz target. These results demonstrate that topology-specific training provides more accurate and stable CSRR designs.",

keywords = "Antennas and propagation, Microwave antennas, Equivalent circuits, Filtering, Circuits, Filters, Protocols, HTTP, Wireless Access in Vehicular Environments, Machine learning, Extreme Gradient Boosting, XGBoost, Complementary Split Ring Resonator, CSRR",

author = "\{De Sousa\}, \{Leonardo O.\} and \{Da Silva\}, \{Mirelli S.\} and Machado, \{Gabriel G.\} and \{De A. Brito Filho\}, Francisco and \{De Holanda\}, \{Samanta M.\} and \{Da Silva\}, \{Isaac B. T.\}",

year = "2026",

month = apr,

day = "13",

doi = "10.1109/lacap67012.2026.11471168",

language = "English",

isbn = "979-8-3315-9780-1",

pages = "1--2",

booktitle = "2026 IEEE 2nd Latin American Conference on Antennas and Propagation (LACAP)",

publisher = "IEEE",

address = "United States",

note = "2026 IEEE 2nd Latin American Conference on Antennas and Propagation , LACAP ; Conference date: 22-02-2026 Through 25-02-2026",

}

De Sousa, LO, Da Silva, MS, Machado, GG, De A. Brito Filho, F, De Holanda, SM & Da Silva, IBT 2026, Evaluating Single- and Multi-Topology XGBoost Models for CSRR Inverse Design. in 2026 IEEE 2nd Latin American Conference on Antennas and Propagation (LACAP). IEEE, pp. 1-2, 2026 IEEE 2nd Latin American Conference on Antennas and Propagation , Natal, Brazil, 22/02/26. https://doi.org/10.1109/lacap67012.2026.11471168

TY - GEN

T1 - Evaluating Single- and Multi-Topology XGBoost Models for CSRR Inverse Design

AU - De Sousa, Leonardo O.

AU - Da Silva, Mirelli S.

AU - Machado, Gabriel G.

AU - De A. Brito Filho, Francisco

AU - De Holanda, Samanta M.

AU - Da Silva, Isaac B. T.

PY - 2026/4/13

Y1 - 2026/4/13

N2 - This work investigates XGBoost-based inverse-design models for circular and square CSRRs. Single-topology models achieved the best performance, with MAEHFSS of 0.0457 and 0.0443 GHz for the circular and square cases, respectively, and low standard deviations across bandwidth variations. In contrast, multi-topology models showed large deviations, with MARHFSS reaching 0.3686 GHz and sensitivity S=1.44. The average S21 responses further confirmed that model C produced resonances closest to the 2.45 GHz target. These results demonstrate that topology-specific training provides more accurate and stable CSRR designs.

AB - This work investigates XGBoost-based inverse-design models for circular and square CSRRs. Single-topology models achieved the best performance, with MAEHFSS of 0.0457 and 0.0443 GHz for the circular and square cases, respectively, and low standard deviations across bandwidth variations. In contrast, multi-topology models showed large deviations, with MARHFSS reaching 0.3686 GHz and sensitivity S=1.44. The average S21 responses further confirmed that model C produced resonances closest to the 2.45 GHz target. These results demonstrate that topology-specific training provides more accurate and stable CSRR designs.

KW - Antennas and propagation

KW - Microwave antennas

KW - Equivalent circuits

KW - Filtering

KW - Circuits

KW - Filters

KW - Protocols

KW - HTTP

KW - Wireless Access in Vehicular Environments

KW - Machine learning

KW - Extreme Gradient Boosting

KW - XGBoost

KW - Complementary Split Ring Resonator

KW - CSRR

UR - https://pure.ulster.ac.uk/en/publications/043f1720-e949-45d2-8f6a-014d066a0058

U2 - 10.1109/lacap67012.2026.11471168

DO - 10.1109/lacap67012.2026.11471168

M3 - Conference contribution

SN - 979-8-3315-9780-1

SP - 1

EP - 2

BT - 2026 IEEE 2nd Latin American Conference on Antennas and Propagation (LACAP)

PB - IEEE

T2 - 2026 IEEE 2nd Latin American Conference on Antennas and Propagation

Y2 - 22 February 2026 through 25 February 2026

ER -

Evaluating Single- and Multi-Topology XGBoost Models for CSRR Inverse Design

Abstract

Conference

Keywords

Access to Document

Other files and links

Fingerprint

Cite this