. "This method allows the best types of data and wavelength ranges to be identified based on their robustness, or lack of propensity to over-train." . . . .