Standard-free calibration transfer - An evaluation of different techniques
|Titel||Standard-free calibration transfer - An evaluation of different techniques|
|Journal||Chemometrics and Intelligent Laboratory Systems|
Within the past decades, the combination of spectroscopic measurements and multivariate calibration techniques has become increasingly applied and a widely acknowledged approach for the extraction of (bio-)chemical information in various applications fields. While obtained prediction models perform satisfyingly in general, the process of data collection, model calibration and model optimization is a time-consuming and cost-intensive one and therefore intended to be repeated as rarely as possible. Unfortunately, changes in the environmental conditions, an adaptation of the measurement setup or an intended or unintended modification of the measured substance itself can all affect spectral measurements and render the calibration model invalid. In such a situation, either a new model needs to be developed or mathematical operations, referred to as (calibration) transfer methods, can be performed to transfer knowledge from the original to the new setting.
In this contribution, we review a large number of transfer approaches available in chemometrics and the field of machine learning, where we focus on techniques applicable in situations where transfer standards, i.e. a set of samples measured under the original as well as the new setting, cannot be provided and only a small amount of reference measurements is available for the new setting. A wide-ranging subset of techniques reviewed in this contribution is also evaluated on industrial data displaying three forms of transfer problems. The superiority of transfer approaches over a plain application of the original model as well as a full model recalibration can be conformed and average rank maps are presented to provide guidance on a proper choice among evaluated techniques.