Machine learning-based tungsten spectroscopy analysis in the WEST tokamak

This article presents a machine learning approach to tungsten spectroscopy analysis using measurements performed in WEST tokamak plasmas. We developed a Random Forest algorithm to predict the maximum electron temperature (Temax) based on 45–65 Å tungsten brightness spectra measured along a mobile line of sight. The model achieves prediction errors typically below 5% across a broad temperature range (0.5–4 keV). Feature importance analysis highlights both physically meaningful wavelengths and others that, despite lacking direct interpretability, improve prediction stability and accuracy. In addition, a principal component analysis is conducted to investigate the relation between spectral variance and Temax unraveling that unaccounted-for parameters influence the spectral shape. Combined with the wavelength importance analysis, this study could potentially inform atomic structure and collisional-radiative models to better understand the tungsten spectral emission.