SHAPCA: Consistent and Interpretable Explanations for Machine Learning Models on Spectroscopy Data — ThinkLLM