Statistical analysis of NIR data: data pretreatment

✍ Scribed by Jianguo Sun

• Publisher: John Wiley and Sons
• Year: 1997
• Tongue: English
• Weight: 117 KB
• Volume: 11
• Category: Article
• ISSN: 0886-9383
• DOI: 10.1002/(sici)1099-128x(199711/12)11:6<525::aid-cem489>3.0.co;2-g

✦ Synopsis

In the statistical analysis of near-infrared (NIR) data arising from the calibration of NIR instruments, two steps are often involved. The first one is data pretreatment, which usually refers to transformation of NIR spectra (e.g. the samples of predictor variables using statistical regression terminology) with the goal of reducing large baseline variations, dimensionality, collinearity and/or noise level of the observed spectra. The pretreatment is needed partly because measured spectra usually have large baseline variation and/or substantial noise and have a low ratio of the sample size to the number of predictor variables. The second step is calibration modeling and involves the application of statistical regression methods to the pretreated NIR data. This paper deals with the data pretreatment step and in particular, a method based on principal component analysis is presented for attacking the problem of large baseline variation. The usefulness of the described method is illustrated through a simulation study and its application to the analysis of a set of real NIR data.