Random Forest Regression-based Model for Fitting Multilayer Isotropic Medium Scattering Problems

In addressing the challenges associated with precision and robustness in both forward and inverse problems, particularly when dealing with data inaccuracies, noise interference, and information distortion, we present an innovative approach. We harness the capabilities of random forests to construct data-driven numerical algorithm models. These decision trees establish high-dimensional mappings in the function space to effectively capture low-dimensional physical information, thus mitigating the impact of noise in the data through the Bagging method. Following the training of the model on the forward problem dataset, it produces results consistent with the accuracy of the finite element method (FEM) in a matter of milliseconds. To validate the efficacy of our model, we applied it to the scattering problem of multi-layer isotropic media, yielding impressive prediction results with an accuracy of 97%. Furthermore, in comparison to traditional numerical algorithms, our method affords grid independence, substantially reducing the challenges associated with grid freedom.