The use of computer modeling techniques in near infrared spectroscopy

Science developments entered a new era in the 21 century. The tendency is clear. The individual scientific fields started to cross each other again in order to have the depth and width in the development. Especially the computer science, it is the fundamental pillar for all the others sciences nowadays. A good example is its applications in Near Infrared spectroscopy (NIR). NIR is a rather new molecular vibrational spectroscopy in the analytical science field. Because of its advantages of being suitable to measure both liquid and solid samples, non destructive to samples, quite acceptable instrumental S/N ratio and sensitivity, NIR technique can be used in many in-line or off-line quantitative analysis. Although NIR only started its development from the 80s of last century, this new technology is developing so quickly that NIR applications have already covered the field of agriculture, material manufacture and chemical industries, petroleum and chemical refining, pharmaceutics and cosmetics etc. It has been widely accepted by many industries, academics and R & D institutes all over the world. However, NIR spectroscopy is a secondary analytical technique. The development of NIR applications is completely dependent upon the good use of Chemometrics, which is a computerized mathematical modeling technique utilized in the analytical field. In the other words, the critical step of a successful NIR application is to make a good NIR calibration model. With a commercially available NIR instrument, to scan the samples and to obtain the NIR spectra are very simple. Furthermore, to use the instrument software to make a pseudo calibration model is also quite easy, but whether the model made will predict the concentration of the relevant components correctly in the unknown samples, is not guaranteed. In this paper, with some real analytical data sets, how to use the major fussy mathematical concepts, such as Principal Components Analysis (PCA), Partial List Square (PLS) etc., in modeling for various applications of NIR spectroscopy will be thoroughly reviewed and discussed.