Feature selection versus feature compression in the building of calibration models from FTIR-spectrophotometry datasets

► We explore the benefit of dimensionality reduction for FTIR datasets. ► We tested the mRMR and SOM dimensionality reduction methods in a quantification task. ► Increasing information will increase the quality of the data if redundancy is low. ► Increasing information will reduce computational costs if redundancy is low. ► Results showed outstanding improvements in quantification performance.