We present a machine learning-enhanced computational framework for predicting the optical properties of two-dimensional silicon arsenide (SiAs). By combining first-principles density functional theory (DFT) calculations with artificial neural networks (ANNs), decision trees (DTs), and random forest regression (RFR), we achieve accurate modeling of both absorption spectra and optical conductivity. Our results demonstrate that RFR delivers the highest quantitative accuracy ( R 2 = 1 . 000 , MAE = 0 . 0005 ), while ANNs provide the most physically realistic continuous spectra. Although DTs provide useful interpretability, their generalization performance is inferior to that of the other approaches. The machine learning models successfully reproduce all key features observed in the DFT calculations, including the prominent absorption peak at 5–6 eV. Detailed analysis of training dynamics reveals that ANNs maintain stable convergence over 500 epochs, while the ensemble approach of RFR effectively compensates for the overfitting tendencies inherent to individual DTs. This hybrid DFT-ML approach provides new insights into SiAs’ optoelectronic properties while establishing a generalizable workflow for accelerating the discovery of 2D materials with tailored optical responses. • A hybrid DFT-machine learning framework is developed for 2D SiAs optical spectra. • ANN, DT, and RFR models are trained to predict absorption and optical conductivity. • RFR achieves the highest accuracy with R² = 1.000 and MAE = 0.0005. • ANN produces smooth and physically realistic optical spectra across 500 epochs. • The DFT-ML workflow accelerates the discovery of 2D materials with tailored optics.
Barhoumi et al. (Sat,) studied this question.