Antennas are fundamental components of modern communication systems, playing a vital role in wireless connectivity and data transmission. However, traditional antenna design and optimization rely heavily on repeated simulations and prototyping, which are both time- consuming and computationally expensive. To address these challenges, this project proposes the development of a machine learning-based model capable of predicting antenna performance parameters—such as gain, return loss, bandwidth, and radiation pattern—based on design and operating conditions. By leveraging artificial intelligence techniques, the proposed approach aims to significantly reduce design time and cost while improving prediction accuracy and efficiency. This research demonstrates the potential of machine learning as a powerful tool for accelerating antenna development and enhancing overall performance prediction capabilities in the rapidly evolving field of wireless communication.

This project demonstrates a Machine Learning approach for predicting antenna performance using Decision Tree Regression. After importing essential libraries and loading the dataset, the code splits the data into training and testing sets. Feature scaling is applied to ensure consistent data representation. The Decision Tree Regressor is trained on the training data and used to predict antenna performance parameters on the test set. Analyzing the visual output of the model’s predictions (yellow line) against the actual values (red line) provides valuable insights. This graphical representation allows for a qualitative assessment of the model’s accuracy. A close match between the predicted and actual values signifies the model’s effectiveness, while deviations indicate areas for potential improvement. To further assess the model, quantitative metrics like Mean Squared Error (MSE) or Root Mean Squared Error (RMSE) can be calculated. Additionally, Decision Tree models are interpretable, enabling a deeper understanding of the underlying rules influencing predictions. By examining the decision rules within the tree structure, domain experts can gain insights into the factors affecting antenna performance predictions. This interpretability is crucial for refining antenna designs based on the model’s findings. Continuous evaluation and optimization are essential. If the model’s performance falls short, further iterations might involve experimenting with different algorithms, exploring advanced feature engineering techniques, or fine-tuning hyperparameters. These iterative refinements are vital in enhancing the accuracy and reliability of the antenna performance predictions, making the project a robust and valuable application of Machine Learning in the domain of antenna engineering.

1.C. Liu et al. [1] introduced advanced machine-learning methodologies for the intelligent design of metamaterials. Their work highlights the effective prediction of S-parameters for coding metamaterials, demonstrating that ML-driven techniques can significantly reduce the complexity of electromagnetic design. 2.L. Zhang et al. [2] proposed a metasurface antenna incorporating dual T-shaped radiating elements, where deep learning is utilized to optimize geometry. Their antenna operates acros 7.9–13 GHz and achieves a peak gain of 16.58 dBi at 13 GHz, showcasing the capability of AI to enhance structural efficiency. 3.J. Nan et al. [3] focused on optimizing MIMO and fractal antenna structures using a hybrid DBN-ELM model. Their model shows strong alignment with target S-parameters and achieves RMSE values of 11.87% and 3.56% for fractal and MIMO antennas respectively, outperforming conventional techniques. 4.N. Kurniawati et al. [4] analyzed multiple ML estimators for antenna parameter prediction. Their study finds that three estimators yield the lowest MAE values for gain prediction, while two exhibit reduced MSE, and the use of eight combined estimators produces the smallest standard error for VSWR analysis. 5.M. Lan et al. [5] introduced a neural autoencoder-based transceiver model that minimizes communication errors. Their approach enhances robustness against channel variations and significantly outperforms traditional modulation and detection schemes in simulations. 6.G. Gampala et al. [6] demonstrated that ML-based surrogate models can replicate original antenna designs with high fidelity. Such models enable designers to run hundreds of optimization cycles within seconds, addressing challenges related to convergence and computational time. 7.In a related study, M. Lan et al. [7] used a deep neural network to model transmitter, channel, and receiver components simultaneously. They employed confidence interval techniques to show the correlation between training sample size and transmission error probability, providing insights into dataset sufficiency. 8.M. Chen et al. [8] discussed the effective creation of training datasets for direction-of-arrival (DOA) estimation. Their auxiliary detection network reduces dataset requirements while maintaining accurate DOA verification after network training. 9.T. Imai et al. [9] demonstrated that CNN-based deep learning models can accurately predict radio propagation characteristics. Their work highlights the efficiency and reliability of deep learning as a replacement for traditional empirical propagation models. 10.M. E. et al. [10] examined how map-derived parameters influence CNN predictions for wireless signal analysis. Their work emphasizes that the abundance of available spatial datasets and the rise in computational power make ML increasingly suitable for RF system optimization 11.R. Tiwari et al. [11] integrated Python Spyder with CST Studio to create a hybrid simulation and ML prediction framework. They showed that Random Forest models achieve 99.56% prediction accuracy, validating their effectiveness in antenna parameter estimation. 12.The authors in [12] confirmed that Random Forest classifiers maintain high accuracy even in high-dimensional electromagnetic datasets, supporting earlier claims of its robustness in handling missing and mixed-type data. 13.S. Kumar et al. [13] demonstrated that convolutional neural networks can learn complex radiation patterns and predict far-field antenna characteristics effectively, reducing the need for full-wave simulations. 14.P. Singh et al. [14] employed gradient-boosted decision trees for predicting return loss and impedance bandwidth, where the model achieved superior performance compared to ANN- based approaches. 15.Roy et al. [15] explored automated antenna tuning using reinforcement learning (RL). Their RL agent autonomously adjusted geometric parameters and achieved faster convergence than manual tuning methods. 16.H. Park et al. [16] presented a GAN-based model that generates synthetic antenna datasets, improving ML performance when real measurement data is limited. 17.Y. Wang et al. [17] utilized support vector regression (SVR) to predict mm-wave antenna gain and radiation efficiency. Their model showed strong generalization even with small datasets, demonstrating ML’s value in early-stage antenna design. 18.F. Ahmed et al. [18] implemented ML-based feature selection techniques to identify the most influential antenna design parameters, reducing computational time while maintaining prediction accuracy. 19.R. Das et al. [19] highlighted that ML accelerates optimization in wearable antennas by predicting bending tolerances and material effects without the need for repeated physical prototyping. S. Patel et al. [20] noted that ML provides automation benefits beyond RF systems, such as in customer-support chatbots. These automation principles translate well into EM engineering, where ML can automate repetitive simulation–optimization cycles.

© 2025 International Journal of Engineering and Techniques (IJET).

Digital Object Identifier (DOI)

IJET assigns a unique DOI to every accepted article via Zenodo for persistent linking and citation. Your article’s DOI: https://doi.org/{{doi}}

Open DOI About Zenodo DOI

Close