Explainable MLP-Based Surrogate Model for Predicting Terahertz Metamaterial Absorber Performance

Abstract

This study presents an explainable surrogate modeling approach for predicting the absorption behavior of broadband THz metamaterial absorbers based on vanadium dioxide (VO2) by a Multi-Layer Perceptron (MLP) network. The proposed model was trained on an electromagnetic full- wave simulation dataset, which included design parameters and frequency, with the corresponding absorption as output. The MLP-based surrogate model was evaluated using performance metrics such as the coefficient of determination (R²), Mean Absolute Error (MAE), and Root Mean Squared Error (RMSE). Additionally, explainable artificial intelligence (XAI) methods were used to understand how the input parameters affect the absorption. Thus, the proposed model offers important clues regarding the physical relationship between design parameters and absorption performance. The results obtained indicate that the MLP model achieves high prediction accuracy. This approach offers a promising alternative for the design and optimization of next-generation THz absorber devices.