A Comparative Analysis of Neural Network Regression Models for Medium Slot Eco-Friendly Microwave Absorber Performance Across Frequency Bands

Accurate prediction of eco-friendly microwave absorber performance across multiple frequency bands is essential for optimizing electromagnetic interference reduction and antenna design. This study presents an artificial neural network (ANN)-based regression model for estimating the absorption performance of a medium slot size eco-friendly microwave absorber across L, S, C and X-bands. A total of 380 experimental data points (25 L-band, 57 S-band, 100 C-band, and 198 X-band) were used after data pre-processing. A feedforward ANN was trained and evaluated using two optimization algorithms: resilient backpropagation (trainrp) and scaled conjugate gradient (trainscg). The best-performing models were configured with hidden layer sizes of 22, 34, 29 for trainrp and 30, 35, 35 for trainscg. Results show that the trainrp algorithm achieved superior performance, obtaining a best validation mean squared error (MSE) of 0.0648 and an overall correlation coefficient R = 0.99431 at only six epochs. In comparison, trainscg achieved an MSE of 0.1598 and R = 0.99331 after 493 epochs. These findings demonstrate that trainrp provides faster convergence and higher generalization accuracy, making it a reliable approach for rapid prediction of eco-friendly absorber performance across wide frequency bands. Future work will focus on extending the model to multilayer and broadband sustainable absorber configurations.