Iodine value determination of edible oils using ATR‐FTIR and chemometric methods

Abstract

Iodine value (IV) was determined using Fourier transform mid infrared spectroscopy with attenuated total reflection and chemometric methods. A set of 245 oil samples were used for calibration purposes, and 76 samples were used for validation purposes. Models developed using full spectra partial least square (PLS), interval PLS (iPLS), forward interval PLS (fiPLS), and backward interval PLS (biPLS) algorithms were compared in terms of performance. Root‐mean‐square error of leave‐one‐out cross‐validation (RMSECV) of 2.68 and root‐mean‐square errors of prediction (RMSEP) of 2.73 were obtained using biPLS algorithm with spectra divided into 34 intervals and combinations of eight intervals (4000–3700, 3600–3400, 3200–3000, 2800–2700, 2300–2100, 1700–1600, 1500–1400, and 1100–1000/cm). This model had a correlation coefficient of 0.9885. In addition, sub‐models based on one oil type had higher performances (RMSECV = 0.24–0.76, RMSEP = 0.29–1.25) than the model based on all oil samples, which suggests that the IV model is oil‐type dependent. However, PLS could compensate for this error provided the calibration set covered enough oil types. The proposed method allowed a rapid determination of IV.Practical applications: IV determination is important for edible oils and oleochemicals. However, traditional wet chemical methods require toxic solvents and reagents. FTIR coupled with multivariate analysis can be used to measure IV. Most of the studies on IV determination using FTIR have focused on specific single oil types without interval selection. A general IV model that incorporates several oil types is expected to improve prediction performances. Therefore, the present method represents a rapid and accurate tool for measuring IV of edible fats and oils.An ATR‐FTIR method combined with interval optimization is described to determine the iodine value in edible oil. The RMSECV for the prediction of IV from isolated wavelength intervals are compared with full‐spectrum PLS model in GA. The result showed that interval optimization is effective in term of improved prediction performance of IV.