NIR Calibration Modeling (Part 2)

( to part 1 )

All the below categories are implemented by using multiple different algorithms and formulas which leads to many different calibrations.

Steps in modeling
  • Data Cleaning – (bad data, missing values, duplicate elimination, spectral quality / intensity / noise, input value typing errors, …)
  • Initial Calibration set up – selection of calibration, validation and test samples
  • Wavelengths selection
  • Data preprocessing, pretreatments
  • Method calculation
  • Choosing the number of Principal Components / Latent Variables
  • Validation of calibration model / Statistics of performance – (accuracy, precision, linearity, repeatability, range, distribution, robustness / stability, sensitivity, simplicity, etc.)
  • Outlier examination and removal

The problem of choosing the optimal number of factors to find the optimum between underfitting and overfitting is solved by having multiple methods and protocols implemented leading to multiple calibrations.

The evaluation and the selection of the best calibration is based on many individual statistical values including the most popular RMSEP, SEP, Bias, SEC, R2 and PCs etc.

Results and Reporting

A detailed calibration report is provided detailing the best available calibration containing all calibration parameter settings and statistics of prediction performance of the calibration set, the validation set and the test set. A visual expression of the calibration is provided with the most importance plots.

Our service works with any quantitative NIR spectra data set in the standard JCAMP-DX format and uses mainly PLS and PCR to be compatible with other chemometric calibration software.