9.3 Reducing Background

Excess variation due to spurious sources can have a detrimental impact on the models (the book mentions principal component regression and partial least square). Removing all the background noise is almost impossible, but we can approximate it.

For spectroscopy data, intensity deviations from zero are called baseline drift and are generally due to noise in the measurement system, interference, or fluorescence (Rinnan, Van Den Berg, and Engelsen 2009) and are not due to the actual chemical substance of the sample.

To get rid of this noise, the author uses a polynomial fit on the lowest intensities. Then they take the negative residuals and subtract those data points.

A polynomial baseline correction for the small-scale bioreactor 1, day 1 intensities.