Schematic of the workflow for training and evaluating ML models to predict excitation temperatures and isomeric ratios from HCN and HNC spectra. Data preparation integrates models and observations to build a dataset. Gaussian fit parameters are used to generate a data cloud of line intensity vs. width, gradient-tagged by temperatures and ratios. Training and testing stages utilise three algorithms, with results benchmarked against radiative transfer models and validated using observational prototype data.