A comparison study of various chemometric approaches, including LS-SVM, for large near-infrared spectroscopic data of feed and feed products found an RMS improvement of 10% to 24%, according to a white paper published in 2011.
This substantial improvement is quite meaningful and will soon be commercially implemented on Bruker’s FT-NIR platforms providing the very best accuracy and stability. Within the food and agriculture industries there are fantastic challenges and rich opportunities for those inclined to apply their professional skills in developing models and computational solutions.
Within these industries, some operations have behaviors that are too complex to be modeled accurately in detail with high confidence. However, some predictive questions more readily lend themselves to study, solution development and solution validation. Such is the case with NIR prediction models for stochastic determination of nutrient values in animal feed and feed ingredients.
Chemometric models such as PLS, MPLS, and ANN have been used for decades. Still, practical everyday use on the plant floor or in the laboratory require these NIR prediction models to be robust and run in the background with little or no operator awareness. With operating margins sometimes as low as 1 or 2% (even negative at times) accuracy counts.
- J.A. Fernández Pierna, B. Lecler, J.P. Conzen, A. Niemoeller, V. Baeten, P. Dardenne, “Comparison of various chemometric approaches for large near infrared spectroscopic data of feed and feed products,” Walloon Agricultural Research Centre (CRA-W), Valorisation of Agricultural Products Department, Food and Feed Quality Unit (U15), Henseval building, Chaussée de Namur 24,5030 Gembloux, Belgium.
- BRUKER OPTIK GmbH, NIR & Process Technology, Rudolf-Plank-Str. 27, 76275 Ettlingen, Germany.