Table 1 Benchmark methods: the outline.

Kriging¹⁰⁹

• Second-order polynomial as a trend function

• Gaussian correlation function

• Hyperparameters found through maximum likelihood optimization

Radial basis functions (RBF)¹⁰⁹

• Gaussian basis functions

• Scaling coefficient adjusted through cross-validation

Gaussian process regression (GPR)¹¹⁰

• Rational Quadratic kernel functions

• Separate GPR models for real and imaginary parts of the response

Hyperparameters optimized through maximum likelihood estimation

Support vector regression (SVR)¹¹¹

• Gaussian (RBF) kernel function

• Separate SVM models for real and imaginary parts of the response

Kernel scale optimized automatically for each frequency

Artificial neural network (ANN 1)¹¹²

• Fully connected architecture with ReLU activation

• Layers: 512 → 256 → 128 → 64 → output (real + imaginary)

Trained using the Adam optimizer with 400 epochs

ANN 2¹¹²

• Fully connected architecture with ReLU activation

• Layers: 128 → 128 → 128 → 128 → 128 → 128 → output (real + imaginary)

• Trained using the Adam optimizer with 400 epochs

ANN 3 ¹¹²

• Fully connected architecture with ReLU activation

• Layers: 256 → 256 → 256 → 256 → 256 → output (real + imaginary)

• Trained using the Adam optimizer with 400 epochs