Machine learning calibration for near infrared spectroscopy data: a visual programming approach

Abstract

Spectroscopy including Near infrared spectroscopy (NIRS) is a non-destructive and rapid technique applied increasingly for food quality evaluation, medical diagnosis, manufacturing, etc. The qualitative or quantitative information using NIRS is only obtained after spectra data calibration process based mathematical knowledge in chemometrics and statistics. This process naturally involves multivariate statistical analysis. Machine learning as a subset of AI (artificial intelligence), in addition to conventional multivariate statistical tools, seems to get more popularity for chemometric calibration of NIRS data nowadays. However, often the software/toolboxes in chemometrics are commercialized version which is not free. For the free versions, programming skills are required to deal with applications of machine learning in spectra data calibration. Therefore, this paper introduces a different approach of spectra data calibration based on visual programming approach using Orange data mining, a free software which is still rarely used by the research community in spectroscopy. The data used namely: pesticide sprayed on cabbage (to classify between pure cabbage and pesticide-sprayed cabbage with different level of pesticide solution), mango sweetness assessment (to predict sugar soluble content in mango based on Brix degree value). These two data represent classification and regression respectively. This approach is intended more for researchers who want to apply machine learning calibration in their spectroscopy data but don’t want to have rigorous programming jobs, i.e. for non-programmers.