Four preprocessing transforms — SNV, MSC, derivatives, and Savitzky-Golay smoothing — decide the quality of a calibration model more often than the choice of regression algorithm itself. This article is a placeholder: fill it in with a description of when each transform helps, and when it removes information the model needs for prediction.
To fill in
- When to use SNV, and when to use MSC.
- The effect of the Savitzky-Golay order and window size on the signal-to-noise ratio.
- The first and second derivative of a spectrum — what they reveal, and what they hide.
- A “before / after” example on a real ScanSpectrum spectrum.
Content in progress.