Durbin-Watson partial least-squares regression applied to MIR data on adulteration with edible oils of different origins

Ozren Jović

Research output: Contribution to journalArticlepeer-review

Abstract

A novel method for quantitative prediction and variable-selection on spectroscopic data, called Durbin-Watson partial least-squares regression (dwPLS), is proposed in this paper. The idea is to inspect serial correlation in infrared data that is known to consist of highly correlated neighbouring variables. The method selects only those variables whose intervals have a lower Durbin-Watson statistic (dw) than a certain optimal cutoff. For each interval, dw is calculated on a vector of regression coefficients. Adulteration of cold-pressed linseed oil (L), a well-known nutrient beneficial to health, is studied in this work by its being mixed with cheaper oils: rapeseed oil (R), sesame oil (Se) and sunflower oil (Su). The samples for each botanical origin of oil vary with respect to producer, content and geographic origin. The results obtained indicate that MIR-ATR, combined with dwPLS could be implemented to quantitative determination of edible-oil adulteration.

Original languageEnglish
Pages (from-to)791-798
Number of pages8
JournalFood Chemistry
Volume213
DOIs
Publication statusPublished - 15 Dec 2016
Externally publishedYes

Keywords

  • Binary mixtures
  • Durbin-Watson statistic
  • MIR
  • Oil adulteration
  • PLS
  • Variable selection

Fingerprint

Dive into the research topics of 'Durbin-Watson partial least-squares regression applied to MIR data on adulteration with edible oils of different origins'. Together they form a unique fingerprint.

Cite this