Transfer Learning Benchmark Dataset
The Melamine dataset comprises Fourier-transform near infrared (FT-NIR) absorbance spectra and turbidity point readings of Melamine Formaldehyde recorded during condensation (i.e. polymerization) at Metadynea GmbH (Krems, Austria). The former are recorded in-line through a fibre optical probe while the latter are measured offline by determination of the turbidity point, i.e. the temperature at which the condensate starts to get turbid. The turbidity point is a measure of polymer length and amount of cross-linking (i.e. the degree of polymerization) and is an important quantity that defines the physical and chemical properties of the final resin.
The dataset comprises recordings on several batches from 4 different recipes that use different raw material ratios and/or additives (see table). The NIR spectra are divided in two spectral ranges 5546 - 6254 and 6596 - 6975 (1/cm) comprising 225 and 121 input variables, respectively. The spectra corresponding to recipes 568 and 861 are shown in the top figure.
| Recipe | # Samples |
|---|---|
| 562 | 3032 |
| 568 | 733 |
| 861 | 3890 |
| 862 | 462 |
The dataset is made public with the owner's explicit consent.
If you use the data please refer to:
R. Nikzad-Langerodi, W. Zellinger, E. Lughofer, S. Saminger-Platz. “Domain-invariant partial-least-squares regression.” Analytical chemistry 90.11 (2018), pp. 6693–6701.
R. Nikzad-Langerodi, W. Zellinger, B. Moser, S. Saminger-Platz. "Domain adaptation for regression under Beer–Lambert’s law." Knowledge-Based Systems 210 (2020), pp. 106447.