Vibrational Spectroscopy 113 (2021) 103222

Contents lists available at ScienceDirect

Vibrational Spectroscopy
journal homepage: www.elsevier.com/locate/vibspec

Recent trends in the use of FTIR spectroscopy integrated with chemometrics

A R T I C L E I N F O A B S T R A C T

Keywords: Edible oils play an essential role in our routine life as cooking or frying oil as well as an ingredient used in food,
Edible oil medicine, and cosmetic commodities. Because of the high consumption demands of edible oils, adulteration
Adulteration incidents have immensely raised. Thus, adulteration detection is very crucial for consumers, oil-producing in­
FTIR spectroscopy
dustries, and regulatory authorities. The traditional analytical methods are usually time-taking, tedious, detri­
Quality control
mental, lacking online monitoring, and need extensive sample preparation. Whereas, the modern oil industry
Chemometrics
needs a competent and non-calamitous analytical approach for the rapid and precise quality control of edible
oils. Fourier transform infrared (FTIR) spectroscopy is an excellent technique for the detection of edible oil
adulteration. It utilizes the fingerprint region of the obtained spectra to differentiate and detect the different
adulterants present in the edible oil. The spectra collected by the FTIR spectroscopy are complex and very much
similar for different edible oils. To solve this complication, the multivariate branch of chemometrics coupled
with FTIR spectral data has been employed to precisely distinguish between different edible oil adulterants. In
chemometrics, different regression models, along with its various robust validation parameters, can detect even
minute adulteration in edible oil with high precision and accuracy. This concise review presents the critical
aspects of major findings and brings together recent studies of the application of FTIR spectroscopy integrated
with chemometric tools from several reliable sources.

1. Introduction The adulteration of edible oils is the prime apprehension for the
consumers and oil processing industries. It includes the replacement of
Edible oils are a vital component of our routine diet by providing expensive food matrices with a cheaper one [5]. This wrong practice of
energy and essential fatty acids, which is crucial for the proper devel­ blending affects the quality of edible oils. Based on lucrative purposes,
opment of the human body [1]. Additionally, edible oils are used as an edible oil adulteration is considered the most sophisticated scam and is
ingredient in pharmaceutical products, cosmetics, and foodstuffs for most regularly forged. There are two principal types of adulterations in
cooking or frying. Edible oils are usually composed of approximately 98 edible oils: the blending of cold-pressed oil with refined oil, the substi­
% triacylglycerols with variations in length, substitution order, and tution of more costly oil with the cheaper one.
degree of saturation of fatty acid chains. The rest of the 2 % consists of Additionally, adulteration of edible oil with oxidized oil is also a
sugars, sterols, carotenes, phospholipids, and lipid soluble-vitamins [2]. major concern for quality control and safety [6]. At the preliminary
Soybean, olive, sunflower, palm, coconut, and rapeseed oils are the most stage, physicochemical properties, including the acid value (AV),
frequently used edible oils these days [3]. The world production of saponification value (SV), iodine value (IV), and fatty acid composition,
soybean oil, rapeseed oil, palm oil, sunflower oil, coconut oil and olive are exploited to discriminate between different edible oils [7]. But most
oil was 56.51, 27.3, 72.27, 21.2, 3.62 and 3.12 million metric tons of these approaches are detrimental, complicated, strenuous, and
respectively in 2019–2020 [4]. time-taking. Additionally, the lack of online monitoring, the use of

* Corresponding author.
E-mail address: [email protected] (D.K. Singh).
1
Rahul Jamwal and Amit have equal contribution and should be treated as first authors.

https://doi.org/10.1016/j.vibspec.2021.103222
Received 6 April 2020; Received in revised form 19 January 2021; Accepted 20 January 2021
Available online 24 January 2021
0924-2031/© 2021 Elsevier B.V. All rights reserved.
R. Jamwal et al. Vibrational Spectroscopy 113 (2021) 103222

noxious chemical reagents, and the production of pernicious waste limit Table 1
the application of these methods in the oils and fat corporations [3]. Absorption peaks of common edible oils in the FTIR spectral region (4000–400
Recently, FTIR, which is one of the prominent vibrational spectros­ cm− 1) along with their assigned functional groups, mode of vibration and in­
copy, has been developed as a promising technique for edible oil tensities [2,3,16].
detection. It is a highly reliable and responsive analytical method for the Wavenumbers Functional groups Mode of vibration Intensities
qualitative as well as quantitative investigation of edible oil adulteration (cm− 1)
[8]. It is a rapid, non-detrimental technique with the least sample 3468 –C––O (ester Overtone Weak
preparation and without the use of perilous chemical reagents or sol­ group)
vents [9,10]. In the last few decades, due to improvements in equipment 3025 –C–H (trans-)
– Stretching Very weak
3008 –C–H (cis-)
– Stretching Weak
and statistical analysis techniques there is an increase in the application 2954 –C–H (–CH3) Stretching Medium
of FTIR spectroscopy. It gives valuable knowledge regarding the (asymmetrical)
composition, structure, and interaction between different functional 2924 –C–H (–CH2) Stretching Very
groups present in the edible oil being investigated [11]. The spectra of (asymmetrical) strong
2852 –C–H (–CH2) Stretching Very
some edible oils appear similar to naked eyes and it makes interpretation
(symmetrical) strong
of the results difficult while in some oils there are significant visible 2730 –C–O
– (ester Fermi resonance Very weak
differences but we cannot quantify the admixtures of two or more oils. group)
Consequently, to solve such issues, the multivariate branch of chemo­ 2678 –C––O (ester Fermi resonance Very weak
metrics is applied to the collected spectral data for the qualitative and group)
1743 –C– (ester Stretching Very
quantitative analysis of edible oil adulteration. This concise review
–O
group) strong
presents the critical aspects of major findings and brings together recent 1711 –C––O (acid) Stretching Very
studies of the application of FTIR spectroscopy integrated with chemo­ strong
metric tools from several reliable sources. 1654 –C–C
– – (cis-) Stretching Very
strong
1648 –C–
–C– (cis-) Stretching Very
2. FTIR spectra acquisition and pre-treatment strong
1465 –C–H (–CH2, Bending (scissor) Medium
FTIR spectral measurements are recorded using an FTIR spectro­ CH3)
photometer with its various accessories. Earlier, transmission FTIR cells 1417 –C–H (cis-)
– Bending (rocking) Weak
1377 –C–H (–CH3) Bending(symmetrical) Medium
have been used as an accessory for edible oil FTIR spectra collection. But 1228 –C–O Stretching Medium
it has many limitations, including difficulty in loading and cleaning of 1155 –C–O, –CH2– Stretching, Bending Strong
these cells, making FTIR cells prone to cross-contamination. Addition­ 1111 –C–O Stretching Medium
ally, in the case of potassium bromide (KBr) pellets, recorded spectra 1098 –C–O Stretching Medium
1033 –C–O Stretching Very weak
may be vulnerable to impurities like water [12]. Along with it, PE
962 –HC– –CH– Out of plane bending Weak
infrared cards, infrared quartz cuvette and Super-resolution imaging (trans-)
(SR) technique (mixing of the oil with a diluent to decrease the viscosity 872 –
–CH2 Bending (wagging) Weak
of oil) have also been used as an accessory with FTIR spectroscopy, but 721 – (CH2)n– Bending (rocking) Medium
the difficulty in the execution of the developed technique r
uled out the application of the above accessories [13,14]. Whereas
chemometrics is applied on the FTIR spectral data for the qualitative and
attenuated total reflectance (ATR) has been an excellent accessory in the
quantitative detection of edible oil adulteration [16].
case of edible oils and fat as it minimizes the difficult sample preparation
without using any perilous chemical reagent [10].
4. Statistical analysis using multivariate chemometrics
3. FTIR spectral analysis of edible oil
The chemical information of oils in the FTIR spectra is present in the
form of bands and intensities, which gives information about the
The mid-infrared spectral region (4000− 400 cm− 1) of the electro­
structure and concentration of the chemical compound. FTIR provides
magnetic spectrum is the most informative region for the analysis of
information from a single measurement as it can record responses at
edible oils having organic compounds as the absorption peaks are
various wavelengths concurrently. That means the information present
because of the vibrations in the specific functional group [2,15]. The use
in the spectra is multivariate thus requires special tools to extract this
of FTIR spectroscopy to edible oils is dependent on the valuable infor­
data into a meaningful form [17]. The most successful approach to
mation hidden within the obtained spectrum in the mid-infrared region.
obtain this analytical information from FTIR data is to use Chemo­
This valuable information is contained in the spectral region called as
metrics. The term chemometrics was derived from the word “kemome­
‘informative’ spectral region. Within the ‘informative’ spectral region of
tri” (Kemo for chemistry and metri for measure) was first coined by
edible oil, 1500–1000 cm− 1 region contains the most important wave­
Svante Wold in 1972 [18]. Chemometrics is an interdisciplinary field
numbers known as the fingerprint spectral region, which can be utilized
that involves mathematical and statistical methods to extract, represent,
to detect the adulteration of edible oils and fats [3]. In the informative
and display maximum chemical information by analyzing chemical data
region, there are many well defined characteristic peaks or bond vi­
[17]. The chemometric analysis involves three main methods, which
brations at particular wavenumber (cm− 1) that are the assignments of
are: Mathematical pre-treatments, Classification methods, and Regres­
the different functional groups present in its fatty acids, as depicted in
sion methods, as shown in Fig. 2.
Table 1. These specific wavenumbers play a crucial role in the qualita­
tive and quantitative analysis of edible oil adulteration. A general FTIR
spectrum of edible oil, such as mustard oil (MO) and its main adulterant 4.1. Mathematical pre-treatments
argemone oil is presented in Fig. 1 [2].
Though FTIR spectra of most of the edible oils seem similar, there are The spectral pre-processing is performed to enhance the signal to
minor differences in the specific band intensities at characteristic noise ratio as FTIR require the co-addition of many scans. Moreover, it
wavenumbers in the spectra of two different edible oils because of the reduces the irrelevant information like scattering effects generated in
variation in their composition [2,5]. These minor differences are not the spectra and, therefore, improves the spectral information. The well-
visually detectable, and to solve this problem, the multivariate branch of known classical pre-treatment methods are smoothening,

2
R. Jamwal et al. Vibrational Spectroscopy 113 (2021) 103222

1
Fig. 1. ATR-FTIR spectra of mustard oil and its adulterated samples with different concentrations of argemone oil in the range of 4000–400 cm− [2].

normalization, baseline corrections, and derivatives [9,19]. within-class variance and maximise separation between classes. The
limitation of the LDA method is that number of variables must be smaller
than the number of observations [16]. On the other hand, SIMCA is
4.2. Classification methods
another commonly used supervised classification method based on PCA
analysis. SIMCA is performed on different classes independently
The multivariate classification includes unsupervised (exploratory
modelled by separate PCA, which explains most of the data variance.
data analysis) and supervised methods. Unsupervised approaches
The number of significant PCs determined for each class is obtained from
attempt to identify the similarities and differences between samples by
the calibration set, whereas the test set is used to predict the unknown
reducing the data dimensionality without any prior information about
samples of all class models. Leave one out cross-validation (LOOCV)
the data. The most widely used unsupervised methods are Principal
method is used within the resulting classes to estimate the capability of
component analysis (PCA) and Hierarchical cluster analysis (HCA) [20].
the model to select the number of PCs. The performance of the SIMCA
PCA is a reduction method that gives a general idea of data and reveals
model depends on the differences between classes but also robustly on
outliers, groups, and trends in the data. PCA works by summarizing the
the calibration set for each class [20,21].
variation in data into a smaller number of new orthogonal variables
called principal components on each of which a score for each sample is
calculated. The score obtained from the PCA is used to see the clustering 4.3. Regression methods
pattern within the data, as similar samples are expected to create cluster
near to each other [21]. Whereas HCA helps in the visualization of The job of regression methods is to relate the spectrum to the
samples relationship based on variable properties and represents its quantifiable properties of the samples. In FTIR for building a multivar­
output in the form of a Dendrogram. Moreover, HCA is based on the iate calibration (MVC), a quantitative linear relationship is developed
fundamental idea of samples being more related to nearby samples between response variables in ‘y’ (concentration) and the predictor
(represented as a cluster) than to samples farther away. variables in ‘X’ (intensities or absorbance) at more than one wavelength
The samples grouped in the same cluster are more similar to each or frequency of a set of known reference samples [2]. Mathematical
other [22]. models were developed by dividing the samples into two subsets,
On the other side, supervised methods are more potent for classifi­ namely the Training set (or calibration set) and the validation set (or
cation studies than unsupervised methods. These methods are also Test set). The training set (X0,y0) is used to build a mathematical model
useful to define calibration and validation sets correctly before the by different regression methods, and further, the model performance is
analysis of multivariate regression. Supervised techniques generate a evaluated or validated through a validation set (X1,y1). The output of the
classification model based on the prior information about the samples model, which is an error of validation includes a related standard error
belonging to a specific class or category [2,22]. Initially, it uses a of validation, is used to configure the model. Both training and valida­
training (calibration) data set to generate a classification model. Then tion sets are used to compute the standard error of calibration (SEC) of
this model is used to predict the class of the unknown samples from an the model. Since every model must be validated in order to test its ability
independent validation set. In some situations, the cross-validation to predict new samples, a prediction set or independent test set must be
technique is used when there is a limited number of a sample present, used [16,21]. Therefore, building a calibration model without validating
which makes it challenging to build an independent validation data set. it makes no sense.
A cross-validation method helps in the validation of model performance The efficiency, accuracy, and precision of the model are usually
by using a training set as a validation set. However, it can generate evaluated by an array of statistical parameters such as Coefficient of
overestimated outcomes [23]. The most commonly used supervised Determination (R2), Root Mean Square Error (RMSE), Relative Predic­
methods are Linear discriminant analysis (LDA), Soft independent tion Error (RE%), and Residual Predictive Deviation (RPD). A superior
modelling by class analogy (SIMCA), and Partial least squares discrim­ model should have higher values of R2 and RPD and lower values of
inant analysis (PLS-DA). A supervised method like LDA is a linear and RMSE and RE %. R2 shows the amount of the explained variance in
parametric method that uses Euclidean distance to minimize either the calibration or prediction set. In contrast, RMSE represents the

3
R. Jamwal et al. Vibrational Spectroscopy 113 (2021) 103222

Fig. 2. Schematic representation showing steps for ATR-FTIR data acquisition for classification and quantification of different edible oil adulterants using che­
mometrics analysis.

estimation of the average uncertainty that can likely be used for the algorithm is used, and it attempts to maximize the covariance within and
prediction of unknown samples. Another important statistical parameter thus capturing the variance and correlating the data together. PLS-R
used to interpret the predictive performance of a model is RPD, which is finds the dimension of maximum variance by constructing new uncor­
the ratio of the standard deviation of the response variable to Root Mean related variables that explain both spectral and concentration data [24,
Square Error of Prediction (RMSEP) and Root Mean Square Error of 25]. Therefore, PLS-R only includes those variables that are significant
Cross-validation (RMSECV) [9,20]. for describing the variations related to the chemical data. PCR and PLS-R
Amid various regression methods available for multivariate cali­ mostly provide similar outcomes, and the relationship among them has
bration, three very powerful factor analysis based methods like Multiple been covered extensively. Both perform data decomposition into spec­
linear regression (MLR), Principal Component Regression (PCR), and tral loadings and scores prior to model building with the aid of these new
Partial Least Squares Regression (PLS-R) are extensively popular in variables. PCR performs decomposition of data by using spectral infor­
chemometrics literature. All methods have been widely used by different mation only whereas, PLS-R applies both spectral and concentration
research groups and successfully applied for quantitative analysis of data. Moreover, PLS-R builds a model by using fewer latent variables
FTIR data. MLR is the oldest regression method and now very less in use than PCR, but it does not affect the predictive ability [18,26].
due to advancements in computer technology [18]. PCR is a widely used The optimum number of latent variables during the development of
method and is divided into two steps. In the first step, the spectral data is multivariate data is vital during regression analysis. If the number of
treated using PCA, which decomposes the spectral data (X) and then variables used increases, then it creates redundancy in the independent
MLR is performed onto the principal component score as predictive variables, which leads to model overfitting. In such situations, the model
variables (concentrations). PCA also suppresses the problem of spectral will offer inferior prediction results and will be dependent greatly on the
co-linearity by omitting vectors having a small magnitude [15,18]. dataset. Contrarily, the use of fewer components provides an underfitted
While PLS is the most widely applied multivariate regression method model, which suggests that the developed model is not powerful enough
and is much related to MLR and PCR, in PLS-R, the least square to detect variability in the dataset [16,27]. Another attribute that creates

4
R. Jamwal et al. Vibrational Spectroscopy 113 (2021) 103222

a problem in the development of a robust model is the presence of quality/priced edible oils such as corn, mustard, sunflower and soybean
outliers, which could occur as a consequence of poor sample processing, oil. A recent work by, Neves & Poppi, used FTIR and data driven soft
instrument malfunctions, interference errors, and poor sensor tuning, independent modeling of class analogy (DD-SIMCA) to check the
etc. A sample is called an outlier according to the relationship with the authenticity and to detect the adulteration of VCO with four low priced
independent variable (spectra) to the dependent variable (concentra­ oil samples (corn, soybean, sunflower and canola) [30]. In this study a
tion) or both variables [27]. In summary, all these applications have total of 304 adulterated samples were prepared in the range 10–40 %
shown that FTIR, along with multivariate analysis, has a vast potential (w/w) for the each adulterant oil. The DD-SIMCA models achieved the
for the quality control of edible oils. acceptable results with the sensitivity ranging from 88.23% to 100.00%
and specificities ranging from of 95.74 to 100.00% for prediction [30].
5. Recent case studies of edible oil adulteration detection Subsequently, Amit et al. [31] demonstrated the suitability FTIR along
with multivariate chemometric modelling for the detection of mustard
In this section, major recent studies of edible oil adulteration oil adulteration in VCO. The LDA analysis achieved 100 % discrimina­
detection using FTIR spectroscopy combined with chemometrics are tion for the calibration and validation sets. Moreover, the authors
summarized’. The main parameters included are edible oil type, adul­ developed multivariate calibration models for determining the concen­
terant, wavenumber region analyzed, and the chemometric method tration of mustard oil in VCO with PCR and PLSR methods. The low
applied. The results of all the major studies are presented in Table 2. values of RMSE (0.125 % v/v) and high R2 value (0.999) obtained
FTIR has been successfully used by various research groups to detect during the models comparison showed that, ATR-FTIR is a fast and cost
edible oil quality changes and adulteration. The technique is generally effective method with the ability be used in monitoring the VCO adul­
coupled along with chemometrics tools such as PCA, LDA, SIMCA, teration [31]. Similarly, using optimized FTIR regions, the PLS-R model
Support vector machine (SVM), K Nearest Neighbor (KNN), PCR, and was developed to detect paraffin oil in the Virgin coconut oil (VCO). The
PLSR, which enables them to understand the chemical and biological blending of harmful paraffin oil causes problems related to the intestine,
information present within the samples. Such a powerful combination of liver, and in extreme cases, even cancer. In this study, the authors used
analytical chemistry and chemometrics and its application to edible oil 128 samples for building PCR and PLS-R models. The best prediction
analysis deliver an approach to expand our knowledge and vision into was obtained for the PLSR model of second derivative spectra at the
edible oils’ inherent chemical composition and to discriminate between optimized region 1800− 700 cm− 1 with high R2 and low RMSE values of
adulterated oil samples accurately. 0.999 and 0.162, respectively [32]. The lowest limit of detection was 1%
Adulteration of expensive oils such as Extra virgin olive oil (EVOO) v/v. The authors suggested that the FTIR spectra around 1800− 700
antedates to the Roman Empire’s time. There are many studies in which cm− 1 optimized wavenumbers could be used as a simple index for
researchers tried to detect adulteration in olive oil. Recently, Vascon­ paraffin adulteration detection in VCO.
celos et al. [15] used FTIR spectroscopy combined with multivariate To demonstrate FTIR’s potential in the discrimination of fake sesame
chemometrics to monitor EVOO adulteration with peanut oil (PEO). The oil products in the Chinese market, a research group tested a set of four
authors used PCA to reduce the spectra to three principal components, different edible oils as adulterants of sesame oil (sesame oil flavor, corn
which explained 92.6 % of the variance. LDA was also applied to classify oil, sunflower oil, and blended oil). The authors found that there was a
the samples using the scores obtained from PCA analysis. The authors peak shift at 1055–1025 cm− 1 with the increase in the concentration of
obtained the highest classification accuracy of 92.3 % for the calibration sesame oil flavor (adulterant). Moreover, SIMCA and PLS-DA results
set and 88.3 % correct classification when cross-validated from LDA. showed efficient discrimination of sesame oil with other fake oils [33].
The PCR and PLS-R models showed high accuracy for prediction with R2 Likewise, Quiñones-Islas et al. [34] presented a study in which SIMCA
values of 0.960 and 0.977, and RMSE of 1.49 and 1.05 % v/v. The study and PLS were applied to detect and quantify avocado oil’s adulteration
highlighted the ability of FTIR to discriminate between adulterated with different edible oils (sunflower, canola, and soybean). The per­
samples with a detection limit of 5% v/v. A similar approach was sub­ centages of adulterants in the avocado oil were evaluated based on the
sequently used to classify 67 olive oil samples from other vegetable chemical parameters such as peroxide value and iodine value. Peroxide
edible oil samples (including soybean, sunflower, rapeseed, corn oil, value is a measure of the extent of hydroperoxides formed through lipid
etc.; n = 79) by Jiménez-Carvelo et al. [28]. The FTIR model developed oxidation during storage of oil. The peroxide values are generally
using PLS-DA and support vector machine classification (SVM-C) was expressed in mmole of peroxide (or active oxygen) per kg of oil (meq O2
able to classify correctly 100 % of the olive oil samples and 92 % of kg− 1). The iodine value indicates the degree of unsaturation in the edible
non-olive oil samples [28]. oil. It is defined as the number of grams of iodine absorbed by 100 g of
Improvements in FTIR spectroscopic instrumentation and chemo­ fat. The higher the iodine value, the more will be the degree of unsa­
metric data analysis technique provide remarkable ability to determine turation, indicating greater exposure to oxidation. The SIMCA model
changes in oil composition due to the addition of harmful or toxic oils provided a 99 % confidence limit and classified 100 % of avocado oil and
that are recognized as detrimental to public health. ATR-FTIR, along adulterants (sunflower oil, canola oil, and soybean oil) correctly. PLS
with PLS-R, has been recently applied for the evaluation of argemone oil detected the adulteration percentage as low as 2% with R2 of greater
in mustard oil. Argemone oil is toxic oil due to the presence of a com­ than 0.98 and SEP in the range of 0.09–2.81 [34].
pound called sanguinarine, which causes epidemic dropsy. It has been The specific fingerprint regions of the mid-infrared region represent
reported that only a 1% concentration of argemone oil is sufficient to the chemical composition and hence have the potential to detect adul­
visualize the symptoms of epidemic dropsy. The authors effectively teration. The adulteration of high priced oil such as Extra virgin olive oil
developed the PLS-R model with a high predictive ability (R2 = 0.999 (EVOO), Red fruit oil (RFO), and Black seed oil (BSO) with cheaper oil or
and RMSE = 0.26), which can detect 1% v/v of the adulterated oil. The less beneficial oil has become an issue of quality and fraud for con­
spectral region 1800–500 cm− 1 from the first derivative of the PLSR sumers. Moreover, differences between various brands of the same oil
model obtained as the best model, and the accuracy and precision of the due to plant origin, variety, and processing conditions, and oil types
model was evaluated by the values of RPD (52.23) and RE % (0.033) having very similar compositions might lead to potential mis­
[29]. classifications of authentic oils. In a study, FTIR in combination with
Recently, many studies have also been carried out using FTIR and multivariate analysis based on PLS was used to distinguish Red fruit oil
chemometrics to detect the adulterants in the Virgin coconut oil (VCO). (RFO) from sunflower oil (SFO) and palm oil (PO) as oil adulterants. The
VCO has gained popularity due to its valuable composition, nutritional, results showed that the frequency region of 1200− 1000 cm− 1 is ideal for
and pharmaceuticals properties. The oil because of these properties is the quantification of SFO in RFO, whereas the region from 1780 to 1680
comparatively expensive and is prone to adulteration with the low cm− 1 is best for the quantification of PO in RFO [35]. Rohman et al. [36]

5
R. Jamwal et al. Vibrational Spectroscopy 113 (2021) 103222

Table 2
Summary of recent adulteration studies of common edible and special oils with their adulterant oils using FTIR spectroscopy coupled with chemometrics.
Edible Oil Adulterant Wavenumber region Chemometrics Results References
selected
1
Mustard oil Argemone oil 3050− 2750 cm− and 1800− 500 cm-1 PCA, LDA, PCR Best prediction detected for 1st derivative spectra [2]
(MO) (AO) and PLS-R and R2 and RMSE values calculated for MO and AO
was 0.999, 0.25 and 0.999 and 0.26 respectively.
Lowest detection limit was 1% v/v.
1
Extra virgin Peanut Oil (PO) 4000− 400 cm− PLS–DA FTIR and PLS-DA was used to identify the [15]
olive oil authenticity of olive oil.
(EVOO)
1 1
Virgin coconut Paraffin oil 3000− 2800 cm− and 1800− 700 cm− PCA, LDA, PCR Best prediction detected for 2nd derivative spectra [16]
oil (PO) and PLS-R and R2 and RMSE values calculated for VCO and PO
(VCO) was 0.999, 0.142 and 0.999 and 0.162 respectively.
Lowest detection limit was 1% v/v.
1 1
Avocado oil Sunflower, canola 4000–550 cm− and 1050–800 cm− SIMCA and PLS PLS model showed R2 value of 0.98, SEC in the [34]
and soybean range of 0.04− 1.47 and SEP between 0.09 and 2.81.
Lowest detection limit detected was as low as 2%.
1
Black cumin Soybean oil (SBO) 4000–650 cm− PCA, SIMCA and R2 and RMSEC for PLSR values were between [10]
seed oil PLSR 0.985− 0.999 and 0.168− 4.476 respectively.
(BCSO) The limit for detection of SBO adulteration in BCSO
was below 5%.
1
Extra virgin Camellia, soybean, 4000–650 cm− PCA, SLLE, LLE SLLE gave satisfactory results. The R2 values ranged [39]
olive oil sunflower and PLS for PLS model were 0.971, 0.998, 0.998 and 0.999.
(EVOO) and corn oils The RMSE values for PLS model were 0.018, 0.095,
0.0172 and 0.022.
1
Sesame Oil Peanut, rapeseed, 3010–2800 and 1800–900 cm− PCA, SIMCA and PLS-DA results showed R2 values of 0.9967 0.9954 [33]
soybean, corn, and PLS-DA 0.9976, 0.9976 and RMSE values of 0.0284 0.0337
sunflower 0.0244 0.0243 for calibration set. The validation
oil. results showed R2 value of 0.9928 0.9848 0.9941
0.9952 and RMSE value of 0.0421 0.0615 0.0381
0.0346 with 100% recognition rate.
Extra virgin Canola oil 3,028–2,985 and 1,200–987 cm-1. DA, PCR and PLS Best results were obtained for normal spectra [40]
olive oil having R2 value of 0.999 and RMSEC value of 0.108.
(EVOO) The validation/prediction results showed R2 of
0.997 and RMSEP of 1.52 % v/v.
Lowest detection level of 1 % v/v of CO in EVOO can
be detected.
Cold Pressed Hazelnut, Canola, 1267− 1209 cm− 1, 1121− 1045 cm− 1, and PCA, HCA, and PLS method determines relationship between actual [41]
Sesame Oil and Sunflower oil 876− 814 cm− 1 PLS adulterant concentration and FTIR predicted
concentrations having R2 value of 0.9633, 0.9526,
and 0.9615.
Olive oil 41 different edible 3588–650 cm− 1(1870–2350 cm− HCA, PCA, and PLS model obtained the R2 of 0.79 and an error [42]
1
(70 different vegetable oil was eliminated) PLS-DA percentage prediction of the model, RMSECV, of
types) 8.28. The method is suitable for screening of edible
oil blends to distinguishing blends with content
lower than 50% (w/w) from those with content
higher than 50% (w/w)
Extra virgin Soybean (SB) and 3,000 and 2,800 cm− 1, 1,730 cm− 1 region PLS with VIP PLS model gave very high correlation between the [43]
olive oil Sunflower (SF) and a third larger region with many scores and PLS- measured and predicted values based on the R2
(EVOO) distinctive wavenumbers between 1,450 and DA (0.996), RMSEC (0.57), RMSECV (0.63), RMSEP
670 cm− 1 (0.41) and bias values (0.006).
PLS-DA model provided 100 % authentication
accuracy of adulteration of SB and DF in EVOO at
weight ratios as low as 1 %.
Olive oil soybean, 3100− 2700 cm–1, 1800− 1600 cm–1 and kNN, SVM-C, The reliability of the different models was [28]
sunflower, 1205− 1080 cm–1 OCPLS, SIMCA, established on the basis of R2, RMSEV, MAEV,
rapeseed, corn oil PLS-R MdAEV. The results obtained (g EVOO/100 g blend)
in terms of R2, RMSEV, MAEV and MdAEV were
0.86, 17.6, 14.6 and 16.0 respectively.
1
Passion fruit oil Sunflower oil 650–4000 cm− PCA and PLS-R PLSR results were evaluated by utilizing different [44]
number of components (n). The norm of residuals
(nor) was used as a measure for the goodness fit
showing best fit for n<4.
Significantly below 1% of the adulterant can be
detected.
1 1
Red fruit oil Sunflower oil 1200–1000 cm− (SFO) 1780–1680 cm− PCA, PLS-R SFO was better determined using FTIR normal [35]
(SFO) and palm oil (PO) spectra at the frequency region of 1200–1000 cm− 1
(PO) with R2 of >0.99 and RMSE of 1.42% v/v
1
Extra virgin Pumpkin seed oil 3020–2995 and 1070–900 cm− LDA, PLS-R Quantification using PLS-R model with R2 of >0.99 [36]
olive oil (PSO) and RMSEC of 0.166 % (vol/vol). LDA successfully
(EVOO) classified EVOO and EVOO blended with PSO.
Black seed oil Walnut oil (WO) 4000–650 cm− 1 (quantification of WO), PLS-R, PCA PLS-R at whole mid infrared region of 4000–650 [37]
(BSO) and sunflower 1724–1203 cm− 1 (Quantification of SFO), cm− 1 is well suited for quantitative analysis of black
oil (SFO) 3009–721 cm− 1(classification). seed oil either in binary mixture or ternary mixture
with walnut oil and sunflower oil.
PLS-R [45]
(continued on next page)

6
R. Jamwal et al. Vibrational Spectroscopy 113 (2021) 103222

Table 2 (continued )
Edible Oil Adulterant Wavenumber region Chemometrics Results References
selected

Avocado oil Soybean oil and 1427–779 cm− 1 (SO in AO), combined Normal spectra were suitable for the quantification
(AO) corn oil regions of 1477–721, 1728–1685, and 3035– of SO in AO which revealed high R2, i.e. 0.9994 and
2881 cm− 1 (CO in AO) low RMSEC, i.e. 0.86% (v/v), Meanwhile, R2 of
0.9994 and RMSEC of 0.87% (v/v) were obtained
by PLS for quantification of CO in AO.
1
Avocado oil Soybean oil (SO) 1500–750 cm− PLS-DA FTIR along with PLS-DA discriminate authentic AO [46]
(AO) and grape seed from the adulterants of Soybean Oil (SO) and Grape
oil (GSO) Seed Oil (GSO) in binary and ternary mixture.
Avocado oil Palm oil (PO) and 1260–900 cm− 1 (AO in PO), the combined PLS-R, PCA PCA is successfully used for looking types of oil [47]
(AO) canola oil region of 3025–2850 having the similarity with AO in terms of FTIR
(CaO) and 1260–900 cm− 1 (AO in CaO) spectra. R2 obtained is > 0.999. RMSEC D
0.80 % (AO in PO) and 0.83 (AO in CaO).
1
Sesame oil (SO) Corn oil (CO) 1072− 935 cm− PLS-R, LDA PLS-R with R2 (0.992) and RMSEP (1.31% v/v) are [48]
obtained. LDA was successfully able to classify
sesame oil and that adulterated with corn oil.
1
Extra virgin Hazelnut oil (HO) 4000–550 cm− CLPP-KNN CLPP accompanied by kNN algorithm was found to [49]
olive oil outperform any other state-of-the-art pattern
(EVOO) recognition techniques.
Extra-virgin Soybean oil and 1900–2600 cm− 1, 3100–4000 cm− 1
PLS-R PLS models can be used to quantify SO and SFO in [50]
flaxseed oil sunflower oil EFO by FTMIR–HATR.
(EVFO)
1
Rosehip oil Soybean, corn, and 3500− 500 cm− PLS–DA MIR and PLS-DA are able to discriminate authentic [51]
(RO) sunflower oils RO from adulterated RO containing soybean, corn
and sunflower oils in different proportions.
Cold-pressed Rapeseed, sesame 3200–2600 cm− 1, 1900–600 cm− 1
PCA, PLS–DA MIR-ATR, combined with dwPLS could be [25]
linseed oil and sunflower oil implemented to quantitative determination of
(CPLO) edible-oil adulteration.

reported the use of FTIR and discriminant analysis (DA) for the detection 6. Conclusion
of adulteration of Extra virgin olive oil (EVOO) with Pumpkin seed oil
(PSO). An accurate classification rate of 100 % was determined for After viewing the research published in the last few years, it is un­
adulterated PSO, with no PSO sample being predicted as authentic. The derstandable that FTIR, coupled with chemometrics, is heading towards
fatty acid composition changes were also used as a criterion to differ­ an exciting direction. This review paper describes the recent advances of
entiate between the two oils. The amount of oleic was decreased with an FTIR along with chemometrics for predicting some major quality pa­
increase in the concentration of PSO in EVOO (R2 of 0.946), while the rameters of edible oils based on multivariate analysis methods such as
amount of linoleic acid was increased with the addition of PSO in EVOO PLS-R, PCR, and MLR. Simultaneous exploration of both FTIR spectra
(R2 of 0.978) [36]. and appropriate chemometrics analysis enables us to characterize oil
FTIR spectroscopic analysis has shown the potential to determine the samples better and consequently improve oil quality and safety status.
binary and ternary mixture of sunflower oil and walnut oil adulteration Thus, FTIR is assessed as a rapid, non-laborious, chemical-free, and
in Black Seed Oil (BSO). Rohman et al. [37] evaluated sunflower oil and sensitive analytical tool for the qualitative detection and safety control
walnut oil adulteration in the concentration range of 0–100 % combined of edible oils.
with the PLS model at the mid-infrared region of 4000–650 cm− 1, pro­
ducing models with R2 values ranging from 0.994 to 0.999 and RMSEC
Declaration of Competing Interest
values ranging from 1.17 to 0.68 % v/v for sunflower oil while for
walnut oil the R2 values ranging from 0.978 to 0.999 and RMSEC values
The authors report no declarations of interest.
ranging from 1.84 to 0.53 % v/v [37]. The authors concluded that using
absorbencies at frequency region of 4000–650 cm− 1, FTIR can be used as
a rapid method to differentiate between the Black Seed Oil and adul­ Acknowledgment
terated samples with sunflower oil and walnut oil at various concen­
trations. Furthermore, Han et al. [38] used FTIR methodology to This work is funded by the International Atomic Energy Agency
determine the possible adulteration of oil (CAO) with six cheaper [22125].
vegetable oils. The binary blends of oils were prepared at various con­
centrations from 1 %–90 % v/v. The authors used ratio of two spectral References
bands (1119 cm− 1 and 1096 cm− 1) to determine adulteration and purity
[1] E. Zahir, R. Saeed, M.A. Hameed, A. Yousuf, Study of physicochemical properties of
of CAO. The calibration models were built according to the spectra ob­
edible oil and evaluation of frying oil quality by Fourier Transform-Infrared (FT-IR)
tained using PLSR algorithm and the performances of models were Spectroscopy, Arab. J. Chem. 10 (2017) S3870–S3876, https://doi.org/10.1016/j.
analysed respectively. The results obtained from the evaluation by arabjc.2014.05.025.
[2] R. Jamwal, Amit, S. Kumari, B. Balan, A.S. Dhaulaniya, S. Kelly, A. Cannavan, D.
RMSEC (1.4518–3.3164 % v/v) and RMSECV (1.7196–3.8136 % v/v)
K. Singh, Attenuated total Reflectance–Fourier transform infrared (ATR–FTIR)
demonstrated that PLSR models can be used to quantitatively detect the spectroscopy coupled with chemometrics for rapid detection of argemone oil
CAO adulteration [38]. adulteration in mustard oil, LWT 120 (2020), 108945, https://doi.org/10.1016/j.
These studies demonstrated the suitability of the FTIR technique as a lwt.2019.108945.
[3] Q. Li, J. Chen, Z. Huyan, Y. Kou, L. Xu, X. Yu, J.M. Gao, Application of Fourier
quality control tool for rapid and swift detection of adulteration in transform infrared spectroscopy for the quality and safety analysis of fats and oils: a
edible oils. Apart from being non-destructive in nature, this technique review, Crit. Rev. Food Sci. Nutr. (2019), https://doi.org/10.1080/
also allows simultaneous measurement of a variety of components. 10408398.2018.1500441.
[4] Statista, Statista - the Statistics Portal for Market Data, Market Research and
Market Studies, Statista, 2019. https://www.statista.com/.
[5] E. Hong, S.Y. Lee, J.Y. Jeong, J.M. Park, B.H. Kim, K. Kwon, H.S. Chun, Modern
analytical methods for the detection of food fraud and adulteration by food

7
R. Jamwal et al. Vibrational Spectroscopy 113 (2021) 103222

category, J. Sci. Food Agric. 97 (2017) 3877–3896, https://doi.org/10.1002/ [29] R. Jamwal, Amit, S. Kumari, B. Balan, A.S. Dhaulaniya, S. Kelly, A. Cannavan, D.
jsfa.8364. K. Singh, Attenuated total Reflectance–Fourier transform infrared (ATR–FTIR)
[6] J. Song, J. Park, J. Jung, C. Lee, S.Y. Gim, H.J. Ka, B.R. Yi, M.J. Kim, C. Il Kim, J. spectroscopy coupled with chemometrics for rapid detection of argemone oil
H. Lee, Analysis of trans fat in edible oils with cooking process, Toxicol. Res. adulteration in mustard oil, LWT 120 (2020), 108945, https://doi.org/10.1016/j.
(2015), https://doi.org/10.5487/TR.2015.31.3.307. lwt.2019.108945.
[7] M.A.P. Muniz, M.N.F. Dos Santos, C.E.F. Da Costa, L. Morais, M.L.N. Lamarão, R. [30] M.D.G. Neves, R.J. Poppi, Authentication and identification of adulterants in virgin
M. Ribeiro-Costa, J.O.C. Silva-Júnior, Physicochemical characterization, fatty acid coconut oil using ATR/FTIR in tandem with DD-SIMCA one class modeling, Talanta
composition, and thermal analysis of Bertholletia excelsa HBK oil, Pharmacogn. 219 (2020), 121338, https://doi.org/10.1016/j.talanta.2020.121338.
Mag. (2015), https://doi.org/10.4103/0973-1296.149730. [31] Amit, R. Jamwal, S. Kumari, A.S. Dhaulaniya, B. Balan, S. Kelly, A. Cannavan, D.
[8] B. Antony, S. Sharma, B.M. Mehta, K. Ratnam, K.D. Aparnathi, Study on FT-MIR K. Singh, Utilizing ATR-FTIR spectroscopy combined with multivariate
spectra of ghee (anhydrous milk fat), Br. Food J. (2017), https://doi.org/10.1108/ chemometric modelling for the swift detection of mustard oil adulteration in virgin
BFJ-06-2016-0276. coconut oil, Vib. Spectrosc. 109 (2020), 103066, https://doi.org/10.1016/j.
[9] J. Andrade, C.G. Pereira, J.C. de Almeida Junior, C.C.R. Viana, L.N. de O. Neves, P. vibspec.2020.103066.
H.F. da Silva, M.J.V. Bell, V. de C. dos Anjos, FTIR-ATR determination of protein [32] Amit, R. Jamwal, S. Kumari, A.S. Dhaulaniya, B. Balan, D.K. Singh, Application of
content to evaluate whey protein concentrate adulteration, LWT 99 (2019) ATR-FTIR spectroscopy along with regression modelling for the detection of
166–172, https://doi.org/10.1016/j.lwt.2018.09.079. adulteration of virgin coconut oil with paraffin oil, LWT 118 (2020), 108754,
[10] F.N. Arslan, G. Akin, Ş.N. Karuk Elmas, I. Yilmaz, H.G. Janssen, A. Kenar, Rapid https://doi.org/10.1016/j.lwt.2019.108754.
detection of authenticity and adulteration of cold pressed black cumin seed oil: a [33] X. Zhao, D. Dong, W. Zheng, L. Jiao, Y. Lang, Discrimination of adulterated sesame
comparative study of ATR–FTIR spectroscopy and synchronous fluorescence with oil using mid-infrared spectroscopy and chemometrics, Food Anal. Methods
multivariate data analysis, Food Control (2019), https://doi.org/10.1016/j. (2015), https://doi.org/10.1007/s12161-015-0125-7.
foodcont.2018.11.055. [34] N. Quiñones-Islas, O.G. Meza-Márquez, G. Osorio-Revilla, T. Gallardo-Velazquez,
[11] A. Rohman, The use of infrared spectroscopy in combination with chemometrics Detection of adulterants in avocado oil by Mid-FTIR spectroscopy and multivariate
for quality control and authentication of edible fats and oils: a review, Appl. analysis, Food Res. Int. 51 (2013) 148–154, https://doi.org/10.1016/j.
Spectrosc. Rev. (2017), https://doi.org/10.1080/05704928.2016.1266493. foodres.2012.11.037.
[12] D. Yamada, H. Kandori, FTIR spectroscopy of flavin-binding photoreceptors. [35] A. Rohman, S. Riyanto, A.M. Sasi, F.M. Yusof, The use of FTIR spectroscopy in
Methods Mol. Biol., 2014, pp. 361–376, https://doi.org/10.1007/978-1-4939- combination with chemometrics for the authentication of red fruit (Pandanus
0452-5_14. conoideus Lam) oil from sunflower and palm oils, Food Biosci. (2014), https://doi.
[13] L. Xu, T. Fei, Q. Li, X. Yu, L. Liu, Qualitative analysis of edible oil oxidation by FTIR org/10.1016/j.fbio.2014.05.007.
spectroscopy using a mesh “cell”, Anal. Methods (2015) https://doi.org/10.1039/ [36] A. Rohman, Y.B. Che Man, A.F. Nurrulhidayah, Fourier-transform infrared spectra
c5ay00438a. combined with chemometrics and fatty acid composition for analysis of pumpkin
[14] Z. An, X. Jiang, G. Xiang, L. Fan, L. He, W. Zhao, A simple and practical method for seed oil blended into olive oil, Int. J. Food Prop. (2015), https://doi.org/10.1080/
determining iodine values of oils and fats by the FTIR spectrometer with an 10942912.2012.654564.
infrared quartz cuvette, Anal. Methods (2017), https://doi.org/10.1039/ [37] A. Rohman, D. Wibowo, Sudjadi, E. Lukitaningsih, A.S. Rosman, Use of fourier
c7ay00727b. transform infrared spectroscopy in combination with partial least square for
[15] M. Vasconcelos, L. Coelho, A. Barros, J.M.M.M. de Almeida, Study of adulteration authentication of black seed oil, Int. J. Food Prop. (2015), https://doi.org/
of extra virgin olive oil with peanut oil using FTIR spectroscopy and chemometrics, 10.1080/10942912.2014.908207.
Cogent Food Agric. 1 (2015) 1–13, https://doi.org/10.1080/ [38] J. Han, R. Sun, X. Zeng, J. Zhang, R. Xing, C. Sun, Y. Chen, Rapid classification and
23311932.2015.1018695. quantification of Camellia (Camellia oleifera Abel.) oil blended with rapeseed oil
[16] Amit, R. Jamwal, S. Kumari, A.S. Dhaulaniya, B. Balan, D.K. Singh, Application of using FTIR-ATR spectroscopy, Molecules. 25 (2020) 2036, https://doi.org/
ATR-FTIR spectroscopy along with regression modelling for the detection of 10.3390/molecules25092036.
adulteration of virgin coconut oil with paraffin oil, LWT 118 (2020), https://doi. [39] X. Sun, W. Lin, X. Li, Q. Shen, H. Luo, Detection and quantification of extra virgin
org/10.1016/j.lwt.2019.108754. olive oil adulteration with edible oils by FT-IR spectroscopy and chemometrics,
[17] E. Domingo, A.A. Tirelli, C.A. Nunes, M.C. Guerreiro, S.M. Pinto, Melamine Anal. Methods 7 (2015) 3939–3945, https://doi.org/10.1039/C5AY00472A.
detection in milk using vibrational spectroscopy and chemometrics analysis: a [40] A. Rohman, Y.B. Che Man, F.M. Yusof, The use of FTIR spectroscopy and
review, Food Res. Int. (2014), https://doi.org/10.1016/j.foodres.2013.11.006. chemometrics for rapid authentication of extra virgin olive oil, JAOCS, J. Am. Oil
[18] A. El-Gindy, G.M. Hadad, Chemometrics in pharmaceutical analysis: an Chem. Soc. 91 (2014) 207–213, https://doi.org/10.1007/s11746-013-2370-5.
introduction, review, and future perspectives, J. AOAC Int. (2012), https://doi. [41] G. Ozulku, R.M. Yildirim, O.S. Toker, S. Karasu, M.Z. Durak, Rapid detection of
org/10.5740/jaoacint.SGE-El-Gindy. adulteration of cold pressed sesame oil adultered with hazelnut, canola, and
[19] D. Granato, P. Putnik, D.B. Kovačević, J.S. Santos, V. Calado, R.S. Rocha, A.G. Da sunflower oils using ATR-FTIR spectroscopy combined with chemometric, Food
Cruz, B. Jarvis, O.Y. Rodionova, A. Pomerantsev, Trends in chemometrics: food Control (2017), https://doi.org/10.1016/j.foodcont.2017.06.034.
authentication, microbiology, and effects of processing, Compr. Rev. Food Sci. [42] P. de la Mata, A. Dominguez-Vidal, J.M. Bosque-Sendra, A. Ruiz-Medina,
Food Saf. 17 (2018) 663–677, https://doi.org/10.1111/1541-4337.12341. L. Cuadros-Rodríguez, M.J. Ayora-Cañada, Olive oil assessment in edible oil blends
[20] A.S. Dhaulaniya, B. Balan, A. Yadav, R. Jamwal, S. Kelly, A. Cannavan, D.K. Singh, by means of ATR-FTIR and chemometrics, Food Control 23 (2012) 449–455,
Development of an FTIR based chemometric model for the qualitative and https://doi.org/10.1016/j.foodcont.2011.08.013.
quantitative evaluation of cane sugar as an added sugar adulterant in apple fruit [43] A. Oussama, F. Elabadi, S. Platikanov, F. Kzaiber, R. Tauler, Detection of olive oil
juices, Food Addit. Contam. - Part A Chem. Anal. Control. Expo. Risk Assess. 37 adulteration using FT-IR spectroscopy and PLS with variable importance of
(2020) 539–551, https://doi.org/10.1080/19440049.2020.1718774. projection (VIP) scores, J. Am. Oil Chem. Soc. 89 (2012) 1807–1812, https://doi.
[21] B. Balan, A.S. Dhaulaniya, R. Jamwal, Amit, K.K. Sodhi, S. Kelly, A. Cannavan, D. org/10.1007/s11746-012-2091-1.
K. Singh, Application of Attenuated Total Reflectance-Fourier Transform Infrared [44] J. Kiefer, A.I. Lampe, S.F. Nicoli, M. Lucarini, A. Durazzo, Identification of passion
(ATR-FTIR) spectroscopy coupled with chemometrics for detection and fruit oil adulteration by chemometric analysis of FTIR spectra, Molecules (2019),
quantification of formalin in cow milk, Vib. Spectrosc. 107 (2020), https://doi.org/ https://doi.org/10.3390/molecules24183219.
10.1016/j.vibspec.2020.103033. [45] F.A. Lumakso, A. Rohman, M. Handoy, S. Riyanto, F.M. Yusuf, Detection and
[22] R.G. Brereton, Pattern recognition in chemometrics, Chemometr. Intell. Lab. Syst. quantification of soybean and corn oils as adulterants in avocado oil using Fourier
149 (2015) 90–96, https://doi.org/10.1016/j.chemolab.2015.06.012. transform mid infrared (FT-MIR) spectroscopy aided with multivariate calibration,
[23] J.M. Bosque-Sendra, L. Cuadros-Rodríguez, C. Ruiz-Samblás, A.P. de la Mata, J. Teknol. (2015), https://doi.org/10.11113/jt.v77.3640.
Combining chromatography and chemometrics for the characterization and [46] A. Rohman, F.A. Lumakso, S. Riyanto, Use of partial least square-discriminant
authentication of fats and oils from triacylglycerol compositional data-A review, analysis combined with mid infrared spectroscopy for avocado oil authentication,
Anal. Chim. Acta (2012), https://doi.org/10.1016/j.aca.2012.02.041. Res. J. Med. Plant. (2016), https://doi.org/10.3923/rjmp.2016.175.180.
[24] A. Gredilla, S. Fdez-Ortiz de Vallejuelo, N. Elejoste, A. de Diego, J.M. Madariaga, [47] A. Rohman, A. Windarsih, S. Riyanto, Sudjadi, S.A. Shuhel Ahmad, A.S. Rosman, F.
Non-destructive Spectroscopy combined with chemometrics as a tool for Green M. Yusoff, Fourier transform infrared spectroscopy combined with multivariate
Chemical Analysis of environmental samples: a review, TrAC - Trends Anal. Chem. calibrations for the authentication of avocado oil, Int. J. Food Prop. (2016),
(2016), https://doi.org/10.1016/j.trac.2015.11.011. https://doi.org/10.1080/10942912.2015.1039029.
[25] O. Jović, Durbin-Watson partial least-squares regression applied to MIR data on [48] N.A. Fadzlillah, Y.B. Che Man, A. Rohman, FTIR spectroscopy combined with
adulteration with edible oils of different origins, Food Chem. (2016), https://doi. chemometric for analysis of sesame oil adulterated with corn oil, Int. J. Food Prop.
org/10.1016/j.foodchem.2016.07.016. 17 (2014) 1275–1282, https://doi.org/10.1080/10942912.2012.689409.
[26] S. Türker-Kaya, C. Huck, A review of mid-infrared and near-infrared imaging: [49] K. Georgouli, J. Martinez Del Rincon, A. Koidis, Continuous statistical modelling
principles, concepts and applications in plant tissue analysis, Molecules 22 (2017) for rapid detection of adulteration of extra virgin olive oil using mid infrared and
168, https://doi.org/10.3390/molecules22010168. Raman spectroscopic data, Food Chem. (2017), https://doi.org/10.1016/j.
[27] J.J. Roberts, D. Cozzolino, An overview on the application of chemometrics in food foodchem.2016.09.011.
science and technology—An approach to quantitative data analysis, Food Anal. [50] L.M. De Souza, F.B. De Santana, L.C. Gontijo, S.J. Mazivila, W. Borges Neto,
Methods (2016), https://doi.org/10.1007/s12161-016-0574-7. Quantification of adulterations in extra virgin flaxseed oil using MIR and PLS, Food
[28] A.M. Jiménez-Carvelo, M.T. Osorio, A. Koidis, A. González-Casado, L. Cuadros- Chem. 182 (2015) 35–40, https://doi.org/10.1016/j.foodchem.2015.02.081.
Rodríguez, Chemometric classification and quantification of olive oil in blends [51] F.B. de Santana, L.C. Gontijo, H. Mitsutake, S.J. Mazivila, L.M. de Souza, W. Borges
with any edible vegetable oils using FTIR-ATR and Raman spectroscopy, LWT 86 Neto, Non-destructive fraud detection in rosehip oil by MIR spectroscopy and
(2017) 174–184, https://doi.org/10.1016/j.lwt.2017.07.050. chemometrics, Food Chem. 209 (2016) 228–233, https://doi.org/10.1016/j.
foodchem.2016.04.051.