Models for the rapid assessment of water and oil content in olive pomace by near‐infrared spectrometry

Abstract

BACKGROUND The measurement of the water and oil content in olive pomace is crucial for controlling the olive‐oil extraction process. The use of near‐infrared (NIR) spectra could allow the measurement of the oil and water content in olive pomace. RESULTS Partial least squares for pomace oil content on a dry basis reached an error of 2.5% (±0.5). Principal component regression for pomace oil content on a wet basis reached an error of 3.7% (±0.5). Both were suitable for quantitative analysis. Principal component regression for pomace water content reached an error of 6.0% (±2.3), suitable for process control. The relationship between ‘ratio of standard deviation of calibration data to standard error of prediction data’ and ‘range of confident prediction error percentage’ was investigated, it results of hyperbolic type, the constant of the hyperbolic equation depends on the product under analysis: for the olive pomace this constant is equal to 45.60 (±1.78). CONCLUSION Near‐infrared analysis confirmed the possibility of determining the oil and water content in the olive pomace, which is important in the olive oil extraction process control. A new algorithm was used, together with standard statistical algorithms, to identify and remove the less useful wavelengths from the model, improving the overall prediction performance. A new parameter (the ‘range of confident prediction error percentage’) has been proposed for estimating the model's prediction error in an objective way. © 2020 Society of Chemical Industry

Use of wavelength interaction terms to improve near infrared spectroscopy models of donkey milk properties

“…For example, from previous works, for kaki fruit the average value of the constant correlating RPD and RANGEPERR%@95% was equal to 36.39 (±3.70) 19 whereas for olive pomace the average value of the constant was 45.60 (±1.78). 18 In the current study the average value of the constant is 45.95 (±0.52) for protein content, 70.35 (±1.32) for lactose content and 73.84 (±0.82) for dry-matter content.…”

“…The proposed ranges of RANGEPERR%@95% are 18 :- > 20%: Poor – not recommended- 10%–20%: Fair – for screening purposes- 5%–10%: Good – for process control- 1%–5%: Excellent – for quantitative analysis- < 1%: Exceptional – useable in any application…”

“…The proposed method of “coefficient of variation algorithm” (CVA) allowed for the identification and subsequent removal of the interfering wavelengths bringing to spectrophotometric models of great predictive performance. 18,19…”

“…The proposed method of "coefficient of variation algorithm" (CVA) allowed for the identification and subsequent removal of the interfering wavelengths bringing to spectrophotometric models of great predictive performance. 18,19 In this particular work, aiming to further improve the model predictive performance, an additional enhancement step is proposed and tested. The method consists of taking into account the measured absorbance for each wavelength and, in addition, also some interaction terms arising from combining of the absorbances of each couple of wavelengths, because doing this, one can expect some enhancement of the model prediction capability.…”