 A new protocol has been developed which uses a combination of Bayesian neural networks, BNNs, chemical group contribution, CGC, methods, and molecular contribution, MC, methods to accurately predict the absorption spectra of individual molecules or mixtures. This protocol requires less than 100 samples for training, compared to other machine learning, ML, methods which need thousands of samples. It also achieves a mean square error of under 2% when predicting the full UV spectrum of a single molecule, while ML models with smiles as input require over 2000 samples to reach similar accuracy. Additionally, this protocol can be used to accurately predict the absorption spectra of mixtures of molecules, even if they have different structures. The protocol's success is attributed to its use of both chemical principles and data-driven tools, making it applicable to a wide range of problems. This article was authored by Ji Mingfan, Qiao Qian, and Xiao Dongzhou. We are article.tv, links in the description below.