Food Anal.

Methods (2018) 11:959–968

https://doi.org/10.1007/s12161-017-1072-2

Honey Adulteration Detection Using Raman Spectroscopy

Mircea Oroian 1 & Sorina Ropciuc 1 & Sergiu Paduret 1

Received: 29 July 2017 / Accepted: 12 October 2017 / Published online: 23 October 2017
# Springer Science+Business Media, LLC 2017

Abstract In this study, the Raman spectroscopy was used to unique sweetening agent that can be used by humans without
detect honey adulterated with fructose (F), glucose (G), processing. It is superior to other sweeteners, such as refined
inverted sugar (IS), hydrolyzed inulin syrup (IN), and malt cane sugar, beet sugar, and maize syrup, being a valuable
must (M). Thus, 56 samples of authentic honeys (acacia, sun- source of rich nutritious compounds, due to its medical bene-
flower, tilia, polyfloral, and honeydew) and 900 adulterated fits and unique flavor (Guelpa et al. 2016; Li et al. 2017; Zhu
samples (with 5, 10, 20, 30, 40, and 50% fructose, glucose, et al. 2010). However, honey can easily be adulterated with
inverted sugar, malt must, and hydrolyzed inulin syrup) were various cheaper sweeteners, such as refined cane sugar beet
analyzed. The classification of honey authenticity has been sugar, high fructose corn syrup, and maltose syrup, resulting
made using the partial least square linear discriminant analysis in higher commercial profits (Li et al. 2012). Although the
(PLS-LDA), and a total accuracy of 96.54% (authentic honey adulteration is done for short-term economic reasons, it can
vs. adulterated honey) was observed, while in the case of damage the interests of both producers and consumers (Chen
adulterated honey, a total accuracy of 90.00% was observed, et al. 2014).
respectively. The determination of the adulterant agent con- Recently, there have been published many papers on the
centration has been made using partial least squares regression identification of adulterated honey using stable carbon isoto-
(PLSR) and principal component regression (PCR) methods. pic ratio mass spectrometry (Çinar et al. 2014; Simsek et al.
The proposed method can be considered easy and rapid for 2012), NIR spectroscopy (Bázár et al. 2016; Gan et al. 2015),
honey adulteration detection to provide continuous in-line hyper-spectral imaging (Shafiee et al. 2016), and high-
information. performance liquid chromatography (Wang et al. 2015).
These methods are time-consuming, expensive, and require
Keywords Honey . Adulteration . Raman spectra . Statistical a high degree of technical knowledge for data interpretation
analysis (Chen et al. 2014).
Raman spectroscopy is a spectroscopic technique
used in condensed matter physics and chemistry to
Introduction study vibration, rotational, and other low-frequency
modes in a system. Raman spectroscopy combines the
Honey, a naturally sweet and levorotatory carbohydrate-rich advantages of infrared spectroscopy with the advantages
product, is produced by honey bees and collected from the of near-IR spectroscopy. The advantages of Raman
nectar of flowers (Chen et al. 2014). Natural bee honey is a spectroscopy are the following: it can be used in solids
and liquids as well; the sample does not require prepa-
ration; the water content does not interfere; it is a non-
* Mircea Oroian destructive method, highly specific to a chemical finger-
[email protected] print of a material; the spectra are quickly acquired
within seconds; the samples can be analyzed through
1
Faculty of Food Engineering, Stefan cel Mare University of Suceava, glass or a polymer packaging; laser light and Raman
Suceava, Romania scattered light can be transmitted by optical fibers over
960 Food Anal. Methods (2018) 11:959–968

long distances for remote analysis; Raman spectra can range of 250–2339 cm−1 and a spectral resolution of
be collected from a very small volume (< 1 μm in 3 cm−1. The samples were placed into a quartz cell with
diameter); and inorganic materials are easier analyzed 1 cm path (the quartz cell is placed into a cuvette hold-
by Raman than by infrared spectroscopy (Owen et al. er) scanned at an increment of 1 cm−1. Integration time
2006). was of 15 s. Before being used, they were warmed up
The aim of this study was to evaluate the potential of to 55 °C to dissolve any crystals and kept in flasks at
Raman spectroscopy to distinguish the authentic honey 30 °C to remove air bubbles that could interfere with
from adulterated one, to detect the adulteration agent spectra studies (Oroian 2015).
(e.g., glucose, fructose, inverted sugar, hydrolyzed inulin
syrup, and malt must), and to determine the adulteration Spectral Data Pre-treatment
agent concentration using statistical analysis (partial
least square linear discriminant analysis—PLS-LDA, In order to optimize the model (PLS-LDA, PCR,
partial least squares regression—PLSR, and principal PLSR), the spectral data have been submitted to pre-
component regression—PCR). treatment such as auto-scaling, mean-centering (MC),
first and second derivatives, multiplicative scattering
correction (MSC), airPLS, and their combination. Of
Materials and Methods all, airPLS combined with auto-scaling was the suitable
model for better classification and prediction, leading to
Samples the best performance with a small number of PLS-LDA,
PCR, and PLSR components. Auto-scaling equalizes
The samples were purchased from local beekeepers from the variances of all variables and gives the same weight to
northeast part of Romania. The honeys (56 samples) were of all the variables (Li et al. 2012).
different botanical origins (acacia, tilia, sunflower, honeydew,
and polyfloral). Before spectral measurement, the honeys Statistical Analysis
were liquefied in a water bath at 55 °C (Oroian 2012). Six
honey samples of each type have been adulterated with fruc- The PLS-LDA classification model, PLSR, and PCR
tose (F), glucose (G), inverted sugar (IS), hydrolyzed inulin were made using Unscramber X 10.1 (Camo, Norway).
syrup (IN), and malt must (M) in different percentages 5, 10, PLS-LDA is an extension of partial least squares
20, 30, 40, and 50%, respectively. (PLS). The PLS finds a set of latent variables (LVs),
The fructose syrup (65 °Brix) has been made using fructose which are the linear combination of original indepen-
(Sigma Aldrich, Germany). The glucose syrup (65 °Brix) has dent variables (x) with the coefficient of loading
been made using glucose (Sigma Aldrich, Germany), but to weights, w. In the PLS-LDA, the response matrix is a
prevent the crystallization process of the solution, the pH was dummy matrix containing class membership information
corrected to 1–2 with citric acid. The inverted sugar (65 °Brix) for each sample (Aliakbarzadeh et al. 2016).
has been made using sucrose (Sigma-Aldrich, Germany) at PLSR model is the most linear multivariate regression tools
pH 1–2 (the solution was corrected with citric acid for the widely applied in spectral analysis being a quick, efficient, and
hydrolysis process). The hydrolyzed inulin syrup (65 °Brix) optimal regression method based on covariance (Chen et al.
has been obtained by boiling inulin (Enzymes & Derivates S. 2014). It is particularly useful when we need to predict a set of
A. Romania) at pH lower than 2 (the solution was corrected dependent variables from a very large set of independent var-
with citric acid for the hydrolysis process). The malt must was iables (e.g., predictors) (Abdi 2003). For a better performance
obtained from a local brewer factory (S.C. Bermas S. A. of the model, a great number of factors is necessary unless the
Romania) and concentrated till 65 °Brix. PLS model complexity and stability are weakened.
The samples have been corrected to 65 °Brix with distilled The PCR is an effective regression model which can extract
water to avoid spectral complications from naturally occurring feature representations from collinear and high-dimension da-
variations in sugar concentration (Li et al. 2012). ta (Yuan et al. 2016). As in the case of PLSR, it is particularly
useful when there is the need to predict a set of dependent
Raman Spectra Acquisition variables from a very large set of independent variables
(e.g., predictors); for a better performance of the model, a
The spectra were recorded using an i-Raman spectrom- great number of factors is necessary unless the PLS model
eter (EZM-A2-785L, B&W TEK Inc., USA) equipped complexity and stability are weakened.
with a fiber-optic Raman probe, a thermoelectric cooled The performance of each model has been evaluated accord-
CCD detector with 2048 pixels and a 785-nm laser with ing to three parameters like the root mean square error of
a maximum output power of 495 mW in the signal cross-validation (RMSECV), the root mean square error of
Food Anal. Methods (2018) 11:959–968 961

prediction (RMSEP), and the regression coefficient. The amplified and influenced negatively the prediction.
RMSECV was calculated using the following Eq. 1: According to the Fig. 1, authentic honey presents some
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi characteristic bands around: 346 cm−1 (endocyclic δ(C-
∑ni¼1
c
ðyest −yref Þ2 C-C) ring mode of glucose covered the band at
RMSECV ¼ ð1Þ 353 cm−1, which originated from the δ(C-C-C) furanoid
nc
form of fructose), 408 cm − 1 (glucose spectrum),
where nc is the number of samples in the calibration set, yref is 498 cm−1 (skeletal stretching), 606 cm−1 (fructose spec-
the reference measurement value of the sample i, and yest is the trum), 681 cm −1 (stretching of CO and CCO, OCO
estimated value for the sample i by the model constructed bending), 793 cm−1 (glucose spectrum), 845 cm−1 (glu-
when the sample I is removed. The numbers of factors were cose spectrum), 1048 cm−1 (the band is thought to be
determined according to the lowest RMSECV. originated from the ν(C-O) vibration of the glucose ring
The RMSEP was calculated using the Eq. 2: seen in glucose, maltose, and sucrose spectra),
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 1054 cm−1 (the vibration might be caused by a major
u n  2
u ∑ p y −y contribution by the bending vibration of C(1)-H and
t i¼1 est;p ref ;p
RMSEP ¼ ð2Þ COH in carbohydrates and a minor contribution by the
np
vibration of C-N bond in proteins and amino acids), and
where np is the number of samples in the prediction set, yref,p is 1238 cm − 1 (vibration of C(6)-OH and C(1)-OH)
the reference measurement value of the prediction sample i, (Corvucci et al. 2015; Li et al. 2017; Özbalci et al.
and yest,p is the estimated value of the model for prediction 2013).
sample i.
Raman Spectra of Adulterated Honey

Different Raman spectra of a polyfloral honey adulter-

Results and Discussion ated with fructose (F), glucose (G), inverted sugar (IS),
and hydrolyzed inulin syrup (IN) in different percent-
Raman Spectra of Authentic Honey ages, 5, 10, 20, 30, 40, and 50%, respectively, are
shown in the Fig. 2.
Raman spectra of a honey analyzed (raw spectra and The honey adulterated with fructose exhibited a
normalized ones) are shown in Fig. 1; each spectrum higher intensity of Raman spectra at 606 cm−1 (see
corresponds to a honey of a different botanical origin. Fig. 2) (skeletal intensity), 750–850 cm−1 (ring vibra-
The spectra are presented between 250 and 2400 cm−1. tions and C-OH stretch), and 1074 cm−1 (C-O-C cyclic
Of the entire Raman spectrum, only the part between alkyl ethers). All these wavenumbers are specific to
250 and 1300 cm−1 has been used for interpretation as fructose (Corvucci et al. 2015; Li et al. 2017; Özbalci
only here the spectral information is relevant. The re- et al. 2013). In contrast with the honey adulterated with
gion between 1300 and 2400 cm−1 has been eliminated fructose, glucose as an adulteration agent registered a
because using the auto-scaling the noise might be decrease in the Raman intensity at 606, 750–850, and

Fig. 1 Honey Raman spectra. Black line—raw spectra, red line—corrected spectra (color figure online)
962 Food Anal. Methods (2018) 11:959–968

1074 cm−1 (Fig. 2), while at wavenumbers specific to Fig. 2 Honey Raman spectra profile adulterated with different agents:„
glucose the intensity increased: 498 cm −1 (skeletal F—fructose, G—glucose, IN—hydrolyzed inulin syrup, IS—inverted
stretching) and 918 cm−1 (CH, COH bend). sugar, M—malt must

In the case of honey adulterated with hydrolyzed inulin

syrup, it can be observed in the first moment that at of small adulteration percentages (e.g., in the case of
606 cm−1, the intensity of the peak is increasing with the inverted sugar where the two components glucose/
increase in the concentration of hydrolyzed inulin syrup fructose are in the same ratio and the addition does
added; this fact is normal because the inulin syrup has a high not influence the system).
concentration of fructose and the 606 cm−1 is characteristic to
fructose. Other bands which are characteristics of hydrolyzed PLS-LDA of Adulterated Honey
inulin syrup presented an increase in their magnitude:
450 cm−1—skeleton vibration, 680 cm−1—deformation vibra- The PLS-LDA model was used for the discrimination of
tions of the ring, 793 cm−1—ring vibration, 1048 cm−1—C-O- honey adulterated (with hydrolyzed inulin syrup, malt
C cyclic in alkyl ethers, and 1253 cm−1—C-O-C cyclic in must, glucose, and inverted sugar, respectively). All
alkyl ethers (Goodacre et al. 2002). the 900 samples were submitted to calibration and
The honey adulterated with inverted sugar did not cross-validation. The results of the model are shown in
show high differences from the authentic samples (glu- the Table 2. In the case of the calibration, a correct
cose and fructose are in the same amount in inverted classification of 92.00% of the samples was observed,
sugar), but there can be observed an increase in the while in the case of the cross-validation, a correct clas-
Raman intensity at 681 cm−1 (stretching of CO and sification of 89.99% of the samples was observed.
CCO, OCO bending)—specific to fructose—and at Having in view the table data, one can notice that some
793 cm−1 (ring vibration)—specific to glucose spectrum samples adulterated with hydrolyzed inulin syrup are
(Corvucci et al. 2015; Li et al. 2017; Özbalci et al. classified as sample adulterated with fructose; this can
2013). be explained by the fact that hydrolyzed inulin syrup
The malt wort spectra are not very well defined be- contains fructose in high concentrations. In the case of
cause at high concentration, the must appears like an honeys adulterated with glucose, some of them are clas-
opaque substance; however, there can be observed an sified as honey adulterated with inverted sugar, this fact
increase in intensity at 470 cm−1 (skeletal vibration) is due to inverted sugar which contains glucose and
specific to maltose. The adulteration of honey with malt fructose in equal concentrations. The best classification
wort can be made only for honeydew honey, because in was observed in the case of malt wort (100%) which
the case of other honeys, color is changing too much has a different composition from the other adulteration
and they can be considered only honeydew honeys; agents and the Raman spectra is very different from
however, this is not a problem since honeydew is in other adulterants.
great demand and the quantities are not very high at
international level.

PLS-LDA—Original and Adulterated Determination of Adulterant Agent Concentration

The PLS-LDA model was used for the discrimination of Honey is a complex food material, and the adulteration agents
authentic and adulterated honey (with hydrolyzed inulin have a high degree of similarity with the authentic samples.
syrup, malt must, glucose, and inverted sugar, respec- For this reason, it is quite difficult to determine the exact
tively). All the 956 samples were submitted to calibra- amount of the adulteration agent added to the authentic sample
tion and cross-validation. The results of the model are based on spectrum. Keeping into account these findings, the
shown in the Table 1. A correct classification of 98.53% concentration of the adulteration agent can be analyzed using
of the samples has been observed in the case of the chemometrics. Two types of linear model: partial least squares
original validation, while in the case of the cross-vali- regression (PLSR) and principal component regression (PCR)
dation, a correct classification of 96.54% of the samples were used to achieve the optimal regression model.
was observed. The reason for which some authentic
samples are classified as adulterated ones is due to hon- Calibration Models
ey composition which is of 70% monosaccharides (e.g.,
glucose and fructose) (Oroian et al. 2015). In the case In this paper, honey samples (acacia, tilia, sunflower,
of adulterated honey classified as authentic one, we polyfloral, and honeydew) adulterated with different percent-
consider that this wrong classification occurs in the case ages of hydrolyzed inulin syrup, glucose, fructose, inverted
Food Anal. Methods (2018) 11:959–968 963
964 Food Anal. Methods (2018) 11:959–968

Fig. 2 (continued.)

sugar, and malt wort were used to build models. Each group build a model and the data were divided into two subsets: one
(based on the adulteration agent and honey type) was used to of them is called calibration set (to build the model) and the

Table 1 Classification of honeys

according to authentic or Calibration Cross-validation
adulteration group using the PLS-
LDA Sample Adulterated Authentic Total % Adulterated Authentic Total %
correct correct

Adulterated 894 6 900 99.33 876 24 900 97.33

Authentic 8 48 56 85.71 9 47 56 83.93
Total 900 56 956 98.53 885 71 956 96.54
Food Anal. Methods (2018) 11:959–968 965

Table 2 Classification of honeys according to the adulteration agent using the PLS-LDA

Calibration Cross-validation

Sample IN M G F IS Total % correct IN M G F IS Total % correct

IN 162 0 0 18 0 180 90.00 162 0 0 24 0 180 90.00

M 0 180 0 0 0 180 100.00 0 180 0 0 0 180 100.00
G 0 0 156 0 24 180 86.67 0 6 150 0 24 180 83.33
F 6 0 0 174 0 180 96.67 12 0 0 168 0 180 93.33
IS 7 0 3 14 156 180 86.67 14 0 4 12 150 180 83.33
Total 175 180 159 206 180 900 92.00 188 186 154 204 174 900 90.00

IN hydrolyzed inulin syrup, M malt must, G glucose, F fructose, IS inverted sugar

Table 3 Statistical parameters of the PLSR model results in the calibration and prediction set

Adulteration agent Honey PCs Calibration set Prediction set

RMSECV R2 RMSEP R2

Fructose Acacia C8 0.001 0.998 0.012 0.972

Honeydew C1 0.001 0.999 0.001 0.996
Polyfloral C8 0.002 0.982 0.013 0.921
Sunflower C2 0.001 0.998 0.010 0.981
Tilia C11 0.002 0.992 0.012 0.911
Fructose All C15 0.011 0.987 0.013 0.984
Glucose Acacia C3 0.001 0.999 0.016 0.931
Honeydew C9 0.003 0.993 0.021 0.961
Polyfloral C5 0.001 0.999 0.005 0.956
Sunflower C10 0.002 0.970 0.015 0.937
Tilia C8 0.001 0.999 0.011 0.979
Glucose All C14 0.008 0.980 0.014 0.937
Inverted sugar Acacia C5 0.001 0.998 0.019 0.910
Honeydew C15 0.001 0.999 0.013 0.984
Polyfloral C14 0.001 0.998 0.003 0.987
Sunflower C7 0.004 0.929 0.019 0.915
Tilia C15 0.001 0.999 0.012 0.954
Inverted sugar All C6 0.010 0.986 0.012 0.982
Hydrolyzed inulin syrup Acacia C10 0.004 0.981 0.015 0.937
Honeydew C14 0.001 0.999 0.011 0.978
Polyfloral C1 0.004 0.971 0.015 0.914
Sunflower C3 0.001 0.993 0.015 0.948
Tilia C7 0.003 0.969 0.014 0.959
Hydrolyzed inulin syrup All C9 0.009 0.985 0.011 0.977
Malt must Acacia C7 0.001 0.999 0.021 0.912
Honeydew C11 0.003 0.994 0.012 0.982
Polyfloral C14 0.001 0.999 0.011 0.978
Sunflower C7 0.002 0.998 0.020 0.903
Tilia C1 0.002 0.998 0.022 0.908
Malt must All C15 0.004 0.991 0.005 0.984
All All C13 0.009 0.983 0.103 0.981
966 Food Anal. Methods (2018) 11:959–968

1.1
1.05
a other one is to test its robustness. The data used for calibration
represented two thirds of the total data, while for prediction,
1
one third of the total data, respectively. The calibration set for
0.95
each adulteration agent and honey type consisted of 29 sam-
0.9
ples, while for the prediction set consisted of 14 samples. The
0.85
models were built as follows: for each adulteration agent and
0.8
honey type (43 samples for each adulteration agent/honey
0.75
type), the same adulteration agent and all honeys (181 sam-
0.7
ples), and all adulteration agents and all honeys (935 samples).
0.65
0.6
0.6 0.7 0.8 0.9 1 1.1
PLSR Model
1.1
b In this study, the number of PCs for each model depending on
1.05
RMSECV values using the cross-validation model has been
1
optimized. Table 3 displays the statistical parameters of PLSR
0.95
model for each honey and adulterant used. It can be observed
0.9
that the number of PCs is not the same for all honeys (a number
0.85
of 15 principal component (PC) has been used, out of which a
0.8
suitable one has been extracted). In the case of the mixed
0.75
groups, it can be observed a high number of PC (15 PC), while
0.7
in the case of individual honey and adulterant agent, the num-
0.65
ber of PC ranged from 2 to 14. Accordingly to the data in the
0.6
0.6 0.7 0.8 0.9 1 1.1 Table 3, there can be observed some differences between the
Fig. 3 Scatters plot of the mixed group models. a PLSR. b PCR. Red RMSECV and RMSEP values, which can be explained by the
line—best linear fit (color figure online) number of data samples used for each type of honey.

Table 4 Statistical parameters of the PCR model results in the calibration and prediction set

Adulteration agent Honey PCs Calibration set Prediction set

RMSECV R2 RMSEP R2

Fructose Acacia C8 0.001 0.998 0.012 0.972

Honeydew C2 0.001 0.999 0.001 0.995
Polyfloral C8 0.003 0.982 0.013 0.971
Sunflower C10 0.002 0.992 0.013 0.975
Tilia C11 0.002 0.992 0.012 0.913
Fructose All C15 0.011 0.987 0.013 0.987
Glucose Acacia C3 0.001 0.999 0.016 0.932
Honeydew C9 0.003 0.993 0.021 0.961
Polyfloral C4 0.001 0.999 0.005 0.986
Sunflower C13 0.001 0.979 0.017 0.928
Tilia C8 0.001 0.999 0.011 0.979
Glucose All C15 0.008 0.980 0.014 0.940
Inverted sugar Acacia C7 0.001 0.999 0.019 0.905
Honeydew C15 0.001 0.999 0.013 0.984
Polyfloral C7 0.001 0.996 0.003 0.977
Sunflower C7 0.004 0.929 0.019 0.915
Tilia C15 0.001 0.999 0.012 0.954
Inverted sugar All C6 0.010 0.986 0.013 0.981
Hydrolyzed inulin syrup Acacia C15 0.003 0.988 0.012 0.982
Honeydew C14 0.001 0.999 0.011 0.978
Polyfloral C1 0.004 0.951 0.015 0.915
Sunflower C3 0.001 0.993 0.015 0.948
Tilia C15 0.003 0.969 0.015 0.957
Hydrolyzed inulin syrup All C12 0.009 0.985 0.011 0.977
Malt must Acacia C7 0.001 0.999 0.021 0.912
Honeydew C1 0.003 0.994 0.012 0.982
Polyfloral C14 0.001 0.999 0.011 0.978
Sunflower C7 0.002 0.998 0.020 0.903
Tilia C14 0.001 0.998 0.021 0.912
Malt must All C15 0.004 0.991 0.006 0.978
All All PC15 0.009 0.983 0.010 0.981
Food Anal. Methods (2018) 11:959–968 967

The experimental versus predicted data of the Raman spectra predicting the concentration of the adulteration agent in
intensity are shown in the Fig. 3. adulterated honey.

PCR Funding Information Mircea Oroian, Sergiu Paduret, and Sorina

Ropciuc have been financed by the National Authority for Scientific
Research and Innovation, CNCS-UEFISCDI, grant number PN-II-RU-
In this study, the number of PCs for each model in TE-2014-4-0110.
function of the RMSECV values using the cross- Compliance with Ethical Standard
validation model has been optimized. The statistical
parameters of PCR model for each honey and adulter-
ant used are shown in the Table 4. In the case of Conflict of Interest Mircea Oroian declares that he has no conflict of
PLSR, the number of PCs is not the same as in the interest. Sorina Ropciuc declares that she has no conflict of interest.
Sergiu Paduret declares that he has no conflict of interest.
case of all honeys ranging from 1 to 15; in the case of
all honeys and all adulterants, the number of PC is Ethical Approval Not applicable.
maximum (a number of 15 PCs have been used, out
of which a suitable one has been extracted). The ex- Research Involving Human Participants and/or Animals Not
perimental versus predicted data of the Raman spectra applicable.
intensity is shown in Fig. 3.
The two types of models have been applied to all honeys Informed Consent Not applicable.
irrespective of the adulterant used (see Tables 3 and 4). It can
be observed that the two models have quite similar RMSE
values and the regression coefficients do not have great dif- References
ferences. In the case of the model applied to all honeys and
adulterants, the RMSE values and regression coefficients are Abdi H (2003) Partial least square (PLS) regression. Encyclopedia of
social sciences research methods. Thousand Oaks (CA): Sage (pp.
better in the case of PCR model. According to the data pre- 792-795). https://doi.org/10.4135/9781412950589.n690
sented in the tables, there can be observed that the PCR model Aliakbarzadeh G, Parastar H, Sereshti H (2016) Crossmark. Chemom
predicts better than PLSR model in the case of individual Intell Lab Syst 158:165–173. https://doi.org/10.1016/j.chemolab.
honeys and adulterants. These results demonstrated that the 2016.09.002
model built by mixed group was not worse than those models Bázár G, Romvári R, Szabó A, Somogyi T, Éles V, Tsenkova R (2016)
NIR detection of honey adulteration reveals differences in water
built by the same botanical origin honey samples, but even spectral pattern. Food Chem 194:873–880. https://doi.org/10.1016/
better due to the fact that the mixed group contained the most j.foodchem.2015.08.092
information on honey and adulterant. It can be concluded that Chen Q, Qi S, Li H, Han X, Ouyang Q, Zhao J (2014) Determination of
the samples consisting of different types of honey may be used rice syrup adulterant concentration in honey using three-
dimensional fluorescence spectra and multivariate calibrations.
to build the model and not to reduce its accuracy. The scattered Spectrochim Acta A Mol Biomol Spectrosc 131:177–182. https://
plots with the experimental and predicted data of Raman in- doi.org/10.1016/j.saa.2014.04.071
tensity are presented in opposition in the Fig. 3; it can be Çinar SB, Ekşi A, Coşkun I (2014) Carbon isotope ratio (13C/12C) of
observed that many points in the prediction set fall to or close pine honey and detection of HFCS adulteration. Food Chem 157:
10–13. https://doi.org/10.1016/j.foodchem.2014.02.006
to the red line (which represents the best linear fit line between
Corvucci F, Nobili L, Melucci D, Grillenzoni FV (2015) The discrimina-
the experimental and predicted data). The data falling on the tion of honey origin using melissopalynology and Raman spectros-
red line shows that the concentration of the adulteration values copy techniques coupled with multivariate analysis. Food Chem
by the Raman intensity spectra is equal to the experimental 169:297–304. https://doi.org/10.1016/j.foodchem.2014.07.122
data. Gan Z, Yang Y, Li J, Wen X, Zhu M, Jiang Y, Ni Y (2015) Using sensor
and spectral analysis to classify botanical origin and determine adul-
teration of raw honey. J Food Eng 178:151–158. https://doi.org/10.
1016/j.jfoodeng.2016.01.016
Conclusions Goodacre R, Radovic BS, Anklam E (2002) Progress toward the rapid
nondestructive assessment of the floral origin of European honey
using dispersive Raman spectroscopy. Appl Spectrosc 56(4):521–
Raman spectroscopy combined with PLS-LDA, PLSR,
527. https://doi.org/10.1366/0003702021954980
and PCR was successfully applied to detect adulterants Guelpa A, Marini F, du Plessis A, Slabbert R, Manley M (2016)
in honey. Raman spectroscopy combined with PLS- Verification of authenticity of South African honey and fraud detec-
LDA seems more successful in detecting malt wort in tion using NIR spectroscopy. Food Control 73:1388–1396. https://
honey than glucose and inverted sugar. The method doi.org/10.1016/j.foodcont.2016.11.002
Li S, Shan Y, Zhu X, Zhang X, Ling G (2012) Detection of honey adul-
applied is simple and efficient, and it is appropriate teration by high fructose corn syrup and maltose syrup using Raman
for field applications as the sample does not need spectroscopy. J Food Compos Anal 28(1):69–74. https://doi.org/10.
pre-processing. The PCR method is a good method of 1016/j.jfca.2012.07.006
968 Food Anal. Methods (2018) 11:959–968

Li S, Zhang X, Shan Y, Su D, Ma Q, Wen R, Li J (2017) Qualitative and 136(3–4):1444–1452. https://doi.org/10.1016/j.foodchem.2012.09.

quantitative detection of honey adulterated with high-fructose corn 064
syrup and maltose syrup by using near-infrared spectroscopy. Food Shafiee S, Polder G, Minaei S, Moghadam-Charkari N, Van Ruth S, Kuś
Chem 218:231–236. https://doi.org/10.1016/j.foodchem.2016.08. PM (2016) Detection of Honey Adulteration using Hyperspectral
105 Imaging. IFAC-PapersOnLine 49(16):311-314. https://doi.org/10.
Oroian M (2012) Physicochemical and rheological properties of 1016/j.ifacol.2016.10.057
Romanian honeys. Food Biophys 7(4):296–307. https://doi.org/10. Simsek A, Bilsel M, Goren AC (2012) 13C/12C pattern of honey from
1007/s11483-012-9268-x Turkey and determination of adulteration in commercially available
Oroian M (2015) Influence of temperature, frequency and moisture con- honey samples using EA-IRMS. Food Chem 130(4):1115–1121.
tent on honey viscoelastic parameters—neural networks and adap- https://doi.org/10.1016/j.foodchem.2011.08.017
tive neuro-fuzzy inference system prediction. LWT Food Sci Wang S, Guo Q, Wang L, Lin L, Shi H, Cao H, Cao B (2015) Detection of
Technol 63(2):1309–1316. https://doi.org/10.1016/j.lwt.2015.04. honey adulteration with starch syrup by high performance liquid
051 chromatography. Food Chem 172:669–674. https://doi.org/10.
Oroian M, Amariei S, Rosu A, Gutt G (2015) Classification of unifloral 1016/j.foodchem.2014.09.044
honeys using multivariate analysis. J Essent Oil Res 27(6):533–544. Yuan X, Huang B, Ge Z, Song Z (2016) Double locally weighted
https://doi.org/10.1080/10412905.2015.1073183 principal component regression for soft sensor with sample
Owen CA, Notingher I, Hill R, Stevens M, Hench LL (2006) Progress in selection under supervised latent structure. Chemom Intell
Raman spectroscopy in the fields of tissue engineering, diagnostics Lab Syst 153:116–125. https://doi.org/10.1016/j.chemolab.
and toxicological testing. J Mater Sci Mater Med 17:1019–1023. 2016.02.014
https://doi.org/10.1007/s10856-006-0438-6 Zhu X, Li S, Shan Y, Zhang Z, Li G, Su D, Liu F (2010) Detection of
Özbalci B, Boyaci IH, Topcu A, Kadilar C, Tamer U (2013) Rapid anal- adulterants such as sweeteners materials in honey using near-
ysis of sugars in honey by processing Raman spectrum using che- infrared spectroscopy and chemometrics. J Food Eng 101(1):92–
mometric methods and artificial neural networks. Food Chem 97. https://doi.org/10.1016/j.jfoodeng.2010.06.014