Meat Science 92 (2012) 506510

Contents lists available at SciVerse ScienceDirect

Meat Science
journal homepage: www.elsevier.com/locate/meatsci

Rapid discrimination of pork in Halal and non-Halal Chinese ham sausages by Fourier
transform infrared (FTIR) spectroscopy and chemometrics
L. Xu a, 1, C.B. Cai b, H.F. Cui a, 1, Z.H. Ye a,, X.P. Yu a,
a
Zhejiang Provincial Key Laboratory of Biometrology and Inspection & Quarantine, College of Life Sciences, China Jiliang University, Xueyuan Street, Xiasha Higher Education District,
Hangzhou 310018, China
b
Department of Chemistry and Life Science, Chuxiong Normal University, Luchengnan Road, Chuxiong 675000, China

a r t i c l e

i n f o

Article history:
Received 19 November 2011
Received in revised form 26 February 2012
Accepted 18 May 2012
Keywords:
Chinese Ham sausage
Halal
FTIR spectroscopy
LS-SVM
PLSDA

a b s t r a c t
Rapid discrimination of pork in Halal and non-Halal Chinese ham sausages was developed by Fourier transform infrared (FTIR) spectrometry combined with chemometrics. Transmittance spectra ranging from 400 to
4000 cm 1 of 73 Halal and 78 non-Halal Chinese ham sausages were measured. Sample preparation involved
nely grinding of samples and formation of KBr disks (under 10 MPa for 5 min). The inuence of data
preprocessing methods including smoothing, taking derivatives and standard normal variate (SNV) on partial
least squares discriminant analysis (PLSDA) and least squares support vector machine (LS-SVM) was investigated. The results indicate removal of spectral background and baseline plays an important role in discrimination. Taking derivatives, SNV can improve classication accuracy and reduce the complexity of PLSDA.
Possibly due to the loss of detailed high-frequency spectral information, smoothing degrades the model
performance. For the best models, the sensitivity and specicity was 0.913 and 0.929 for PLSDA with SNV
spectra, 0.957 and 0.929 for LS-SVM with second derivative spectra, respectively.
2012 Elsevier Ltd. All rights reserved.

1. Introduction
The concern of food authenticity and adulteration has resulted in
increased awareness regarding the composition of food products.
The identity of the ingredients in processed or composite mixtures
is not always readily apparent. Therefore, verication that the components are authentic and from sources acceptable to special consumers
may be required (Lockley & Bardsley, 2000). Meat in particular is a
medium rich in social meaning because of its association with cultural
habits and rituals, both religious and secular. Religious food prescriptions are far easier to adopt than to discard because once a ban is
adopted it tends to be reinforced by strong feelings of disgust for
example the strong aversion of Muslims for pork in general. Meat
species identication and Halal authentication are a major concern
in Asia, France, Russia, Sweden, Germany, Switzerland, Greece,
Spain, Italy, United Kingdom, South and North America and most
other countries (Murugaiah et al., 2009).
Rapid and reliable methods for detection of Halal meat adulteration are indispensable for implementation of food labeling regulations and product quality control. Methods for these purposes need
to be specic, sensitive, rapid, economic and able to analyze samples
of different morphological characteristics (Meza-Mrquez, Gallardo-

Corresponding authors.
E-mail addresses: zhye@cjlu.edu.cn (Z.H. Ye), yxp@cjlu.edu.cn (X.P. Yu).
1
These authors contributed equally to the work.
0309-1740/$ see front matter 2012 Elsevier Ltd. All rights reserved.
doi:10.1016/j.meatsci.2012.05.019

Velzquez, & Osorio-Revilla, 2010). Various techniques have been

proposed for the analysis of pork or lard, including differential scanning calorimetry (Coni, Pasquale, Cappolelli, & Bocca, 1994;
Kowalski, 1989), gas chromatography (Farag, Abo-raya, Ahmed,
Hewedi, & Khalifa, 1983), high pressure liquid chromatography
(Marikkar, Ghazali, Che Man, Peiris, & Lai, 2005; Rashood, Shaaban,
Moety, & Rauf, 1995; Saeed, Ali, Rahman, & Sawaya, 1989), electronic
nose (Che Man, Gan, NorAini, Nazimah, & Tan, 2005), and DNA-based
methods (Aida, Che Man, Raha, & Son, 2007; Aida, Che Man, Wong,
Raha, & Son, 2005).
Ham sausage accounts for nearly one third of the total meat products in China and the yearly output exceeds 10,000,000 tons. Chinese
ham sausage is a complex mixture consisting mainly of meat and
starch and with low concentrations of water, vegetable oil, salt,
monosodium glutamate and other food additives. The main meat contents of Halal ham sausage are beef, chicken or sh. Unfortunately,
some food manufacturers choose to use pork as a substitute ingredient for Halal meats because it is cheaper and easily available, which
would trigger serious dispute on national relationships. Minced
meat production removes the morphological characteristics of muscle, making it difcult to identify one type of muscle from another.
For this reason, meat substitution with unspecied species, usually
of lower quality, is the most common form of economic adulteration
in the minced meat industry (Hargin, 1996). For routine analysis of
ham sausage, some of the above methods are too laborious, timeconsuming and expensive, therefore, rapid and economical yet reliable analysis methods are highly demanded.

L. Xu et al. / Meat Science 92 (2012) 506510

As a promising alternative approach to the traditional methods of

chemical and sensory analysis, the combination of spectrometry and
chemometric methods has been successfully used in Halal food analysis.
Previous research has shown the potential of FTIR spectroscopy for
analysis of lard in cake formulation (Syahariza, Che Man, Selamat, &
Bakar, 2005) and chocolate products (Che Man, Syahariza, Mirghani,
Jinap, & Bakar, 2005). FTIR spectroscopy was also used to characterize
lard and other edible oils (Guillen & Cabo, 1997). Other successful applications include detection of lard in the mixture with other animal fats
(Che Man & Mirghani, 2001; Jaswir, Mirghani, Hassan, & Mohd Said,
2003; Rohman & Che Man, 2010) and analysis of pork adulteration in
beef meatball (Rohman, Sismindari, Erwanto, & Che Man, 2011). The
objective of this study is to develop an accurate and reliable method
to discriminate Halal and non-Halal Chinese ham sausages by FTIR spectrometry combined with multivariate discriminant analysis. Given the
performances of different classication models and preprocessing
methods are similar or have no signicant differences, models with
least complexity and preprocessing were sought to ensure the generalization of models.
2. Materials and methods
2.1. Collection of samples
Representative Halal and non-Halal ham sausage samples by main
producers from China were collected. 73 Halal and 78 non-Halal ham
sausage samples were analyzed. The ham sausage samples were
obtained from the domestic markets and the identities of samples
were ascertained by the quality branch of manufacturers. All of the
samples retained integral packaging and the original labels indicating
detailed sample information. For all the samples, according to national and professional standards, the content of starch was less than 10%.
The contents of protein and fat were no less than 12% and no more
than 10%, respectively. The detailed information concerning samples
is shown in Table 1. All of the samples were stored at about 0 C
with integral packaging before spectrometry analysis.
2.2. Sample preparation and FTIR spectroscopy analysis
Sample preparation involved nely grinding of ham sausage samples followed by preparation of KBr pellets. Sampling was performed
on different parts of each ham sausage to consider the potential heterogeneity of materials. Then the granular samples were manually
ground into ne particles with KBr using an agate pestle and mortar.
25 mg (1:40 w/w) of each powder sample was mixed with 975 mg
(39:40 w/w) of KBr (Xi'an Shiji, Xi'an, China). KBr pellets were prepared by exerting a pressure of 10 MPa for approximately 5 min in a
pellet press (Tuopu Instrument., Tianjin, China). To examine whether

Table 1
Detailed information of the ham sausage samples analyzed.
Batch size

Meat content

Typesa

Jinluo
Jinluo
Jinluo
Jinluo
Shineway
Shineway
Shineway
Shineway
Yurun
Yurun
Yurun
TRS
Meihao
Meihao

12
11
10
11
10
12
15
14
12
11
8
11
7
7

Pork
Chicken
Beef
Chicken and Pork
Pork
Chicken and Pork
Beef
Chicken
Beef
Chicken
Pork
Pork
Pork
Chicken and Pork

N
H
H
N
N
N
H
H
H
H
N
N
N
N

N = non-Halal and H = Halal.

the variation in pellet thickness cause signicant interference in the

measured spectra, different pellets were prepared from the same
sample and their FTIR spectra were compared (Garip, Gozen, &
Severcan, 2009). The correlation coefcient between each spectrum
with the average spectrum was higher than 0.99, indicating the measured FTIR spectra were nearly identical to their average spectrum
used for analysis.
FTIR spectra were collected using an Avatar-360 FTIR spectrometer (Thermo Scientic, Waltham, MA) working in the wavelength
range of 4004000 cm 1. For each pellet, 128 scans were performed
with a resolution of 4 cm 1 at room temperature using OMNIC software. An increase in scanning time did not signicantly improve the
signal. The average of the 128 scans was used as a raw spectrum
for further data analysis. The scanning interval was 1.929 cm 1.
Therefore, each spectrum contained 1868 individual points for
chemometric analysis.
2.3. Preprocessing and data splitting
All the preprocessing and further data analysis were performed on
Matlab 7.0.1 (Mathworks, Sherborn, MA). Considering the lack of sufcient prior information concerning the measured spectra, different
options were investigated to optimize data preprocessing. Smoothing
can remove part of the random noise present in the signal and enhance the signal-to-noise ratio (SNR). The algorithm of polynomial
tting (Savitzky & Golay, 1964) was adopted for this purpose because
of its popularity and simplicity. Taking derivatives can enhance spectral differences and remove baseline and background, so rst and second derivatives were applied. Because direct differencing tends to
decrease the SNR by enhancing noise, the derivative spectra were
also computed by polynomial tting algorithms. Standard normal
variate (SNV) (Barnes, Dhanoa, & Lister, 1989) transforms each measured spectrum into a signal with zero mean and unit variance. It was
originally proposed to reduce scattering effects in the spectra but was
also proved to be effective in correcting the interference caused by
variations in pellet thickness or optical path. Although the inuence
of the thickness of pellets was found to be insignicant in this work,
SNV was performed to reduce the possible variations caused by scattering effects or uneven mixing of KBr and sample powders.
The DUPLEX algorithm (Snee, 1977) was used to split the measured spectra data into a representative training set and test set.
DUPLEX selects the two samples with largest distance and puts
them in the training set, then selects the two samples with largest
distance among the left samples and puts them in the training set,
and so on. By alternatively selecting the spectral data for the calibration set and test set, DUPLEX gives data in the test set with a distribution almost equal to that of the training set. Because the distributions
of Halal and non-Halal samples were different, DUPLEX method was
performed separately for the Halal and non-Halal ham sausages.
2.4. Multivariate statistical analysis and method validation

Brands

507

Partial least-squares discriminant analysis (PLSDA) (Barker &

Rayens, 2003) is a classication method based on partial least squares
(PLS) regression. As the cornerstone of chemometrics, PLS has been
successfully used to solve various regression problems. For two-class
problems, suppose an n p matrix X including p wavelength variables
for n training objects, the response vector y (n 1) is constructed with
the category variable of each object in X, for instance, +1 and 1
were used to denote Halal and non-Halal samples, respectively. Therefore, the cut-off value of predicted response values was set to be 0,
e.g., an object with a predicted response value above/under 0 would
be classied into Halal/non-Halal class.
Support vector machine (SVM) considers the trade-off between
the capacity and generalization performance of a learned model by
regression coefcients regularization. Based on the structural risk

508

L. Xu et al. / Meat Science 92 (2012) 506510

minimization principle, SVM has been proved to be an effective and

robust method for both classication and regression. Least squares
SVM (LS-SVM) (Suykens & Vandewalle, 1999) was suggested as a
simplied version of SVM. With equality constraints in the formulation, LS-SVM obtains the solution by solving a set of linear equations,
instead of quadratic programming for traditional SVM algorithms. LSSVM can model nonlinear relationship by using a kernel matrix transformation of the original X.
In this paper, linear PLSDA and nonlinear LS-SVM were applied to the
discrimination of non-Halal and Halal ham sausages based on the raw
and preprocessed spectra. For LS-SVM, the mostly frequently used
Gaussian radical basis function was applied for nonlinear transformation.
To reduce the risk of overtting, Monte Carlo cross validation (MCCV)
(Xu & Liang, 2001) was used to evaluate the number of PLSDA latent variables and optimize the parameters of LS-SVM. The mean err rate of
MCCV (MERMCCV) was used as the validity criteria for the classication.
2.5. Evaluation of model performance
Sensitivity and specicity (Forina, Armanino, Leardi, & Drava, 1991)
were used to evaluate the performance of different classication models
and data preprocessing methods. Denote Halal as positive and nonHalal as negative, sensitivity (Sens) and specicity (Spec) were computed as:
TP
TP FN
TN
Spec
TN FP
Sens

where TP, FN, TN, and FP denote the numbers of true positives, false
negatives, true negatives, and false positives, respectively.
Moreover, the total accuracy of classication was also used:
Accu

TN TP
TN TP FN FP

Fig. 2. PCA plot of raw FTIR spectra of Halal and non-Halal ham sausages.

Peaks in 30002800 cm 1 also involve the contributions of C-H

stretching vibrations. Other obvious bands include 1720 cm 1 (C O
stretching) 1530 cm 1 (asymmetric stretching vibration of COOH in
amino acids) and 1070 cm 1 (stretching vibration of COC). Accurate
assignments of most peaks were difcult due to low resolution and signicant baselines, therefore, pattern recognition methods are necessary
to extract the useful information from spectral data for discrimination.
Fig. 2 demonstrates the PCA plot of raw FTIR spectra of Halal and
non-Halal ham sausages. The rst two principal components (PCs)
explain 83.2% of the total variances. It can be seen that the projections
of the samples in both classes onto the 2-PC subspace are very disperse. This can be attributed to the fact that both Halal and nonHalal groups were known to contain samples composed of different
meats. Obviously, the data have a complex structure and two PCs
are insufcient to discriminate them from each other.

3. Results and discussions

3.1. FTIR spectral analysis
Some of the raw FTIR spectra of Halal and non-Halal ham sausages
were shown in Fig. 1. Seen from Fig. 1, the spectra of Halal and nonHalal samples have very similar absorbance bands in the range of
4004000 cm 1 (Ripoche & Guillard, 2001). The wide bands in
36001700 cm 1 can be attributed to the overlapping of the
stretching of various OH groups (36002000 cm 1) and NH
groups (34001700 cm 1), where the peak resolution is very low.

Fig. 1. Raw FTIR spectra of Halal (solid line) and non-Halal (dotted line) ham sausages.

Fig. 3. Average spectra of Halal and non-Halal ham sausages preprocessed by smoothing
(1), SNV (2), rst-order derivative (3) and second derivative (4).

L. Xu et al. / Meat Science 92 (2012) 506510

Fig. 3 demonstrates the average preprocessed spectra of Halal and

non-Halal ham sausage samples. By comparison of the raw spectra
with smoothed spectra, although smoothed spectra can slightly improve the SNR, it might lose some useful high-frequency information
in the raw data. Compared with rst-order derivative spectra, the
second-order derivative spectra can remove most of the baselines
and enhance some detailed information and peak resolution. SNV
spectra can remove some spectral variations while enhancing others.
The effects of data preprocessing should be evaluated by model
performance.
3.2. Optimization of model parameters
The DUPLEX method was performed on the Halal and non-Halal
samples separately. Each data set was divided into a training set of
50 samples and a test set containing the remainder samples. The
two training sets were then combined for developing classication
models. Therefore, the nal training set had 100 objects (50 Halal
plus 50 non-Halal) and 51 test samples (23 Halal plus 28 non-Halal).
LS-SVM and PLSDA models were developed based on raw and
preprocessed spectra. For LS-SVM, two parameters, and need to
be optimized. The kernel width parameter, , is related to the condence in the data; the magnitude of also inuences the non-linear
nature of the regression. As decreases, the kernel becomes
narrower, forcing the model toward a more complex (nonlinear) solution. The regularization parameter controls the tradeoff between
maximizing the margin and minimizing the training error, a too
small value of will lead to an under-tted model; if is too large,
the model tends to overt the training data and the model will
have poor prediction performance. Therefore, should be optimized
together with the kernel width parameter . To optimize the two
parameters in LS-SVM and the model complexity (number of latent
variables) of PLSDA, MERMCCV was computed with different combinations of parameters. Compared with the traditional leave-one-out
cross validation (LOOCV), MCCV can effectively reduce the risk of
overtting by multiple resampling of the training set and a higher
rate of leave-out samples. In this paper the resampling time is set to
be 200 and the number of left-out samples was 10 (5 Halal and 5
non-Halal) for each resampling. The MERMCCV was computed as:

ERMCCV

N
X
ENi
NL
i1

509

Table 3
Results of LS-SVM models with different preprocessing methods.
Preprocessing

Sensitivity

Raw data
Smoothing
1st derivative
2nd derivative
SNV

0.826
0.783
0.870
0.957
0.913

a
b

(19/23)a
(18/23)
(20/23)
(22/23)
(21/23)

Specicity
0.893
0.893
0.786
0.929
0.929

(25/28)b
(25/28)
(22/28)
(26/28)
(26/28)

2,
0.80,
0.65,
0.85,
0.45,
0.40,

MERMCCV Accuracy
8
11
14
7
11

0.170
0.191
0.211
0.091
0.094

0.824
0.784
0.784
0.922
0.902

True positive/Total positive.

True negative/Total negative.

that preprocessing generally improved the classication performance

in terms of sensitivity and specicity. However, PLSDA and LS-SVM
based on smoothed spectra had inferior performance, which might
be attributed to the possible loss of detailed frequency information
(Kokalj, Rihtari, & Kreft, 2011). Second derivative and SNV signicantly sharpened the classication models by reducing the baseline
and backgrounds. The model complexity of PLSDA based on rst derivative, second derivative and SNV was reduced compared with the
model based on smoothed and raw spectra. The results by rst derivative spectra were not satisfying and this might be partially attributed
to the baseline remained in rst derivative spectra as seen in Fig. 3.
For the best models, the sensitivity and specicity was 0.913 and
0.929 for PLSDA with SNV spectra and 0.957 and 0.929 for LS-SVM
with second derivative spectra, respectively. The best prediction results were also demonstrated in Fig. 4. The comparison of different
preprocessing methods demonstrated that the spectral variations
caused by scattering effects and baseline shifts played a more important role than SNR. Since LS-SVM involves nonlinear transformation
of the raw variables, PLSDA with SNV preprocessing should be

where N is the number of resampling time, L is the number of leaveout samples, ENi is the misclassied samples for the ith resampling.
The number of PLSDA latent variables and LS-SVM parameter combination ( and ) were optimized to obtain the lowest MERMCCV
values.
3.3. Comparison of model performances
With different preprocessing methods, the prediction results and
optimized parameters were listed in Tables 2 and 3. It can be seen

Table 2
Results of PLSDA models with different preprocessing methods.
Preprocessing

Sensitivity

Raw data
Smoothing
1st derivative
2nd derivative
SNV

0.783
0.739
0.826
0.913
0.913

a
b
c

(18/23)b
(17/23)
(19/23)
(21/23)
(21/23)

Specicity
0.857
0.821
0.750
0.893
0.929

The number of PLS latent variables.

True positive/Total positive.
True negative/Total negative.

(24/28)c
(23/28)
(21/28)
(25/28)
(26/28)

Lva

MERMCCV

Accuracy

8
7
6
6
7

0.170
0.191
0.211
0.102
0.091

0.824
0.784
0.784
0.922
0.902
Fig. 4. The predicted response values by the best linear PLSDA model and nonlinear LSSVM models. Samples 123 are Halal (positive) samples and 2451 are non-Halal
(negative) samples.

510

L. Xu et al. / Meat Science 92 (2012) 506510

recommended because it is linear and simpler and expected to have a

more reliable generalization performance.
4. Conclusion
Rapid discrimination of Halal/non-Halal Chinese ham sausages
was developed by FTIR spectroscopy and chemometric data analysis.
PLSDA with SNV spectra (sensitivity 0.913 and specicity 0.929)
and LS-SVM with second derivative spectra (sensitivity 0.957 and
specicity 0.929) achieved best classication performance in terms
of prediction sensitivity and specicity. The analysis results indicate
removal of spectral background and baseline plays a more important
role than a higher signal-to-noise ratio (SNR). Taking derivatives, SNV
can not only improve classication accuracy but also reduce the complexity of PLSDA. Possibly due to the loss of detailed high-frequency
spectral information, spectra smoothing degrades the model performance. Although we can hardly perform an exhaustive sampling of
all types of ham sausages, this study built a reliable model for Halal/
non-Halal discrimination of some mainstream and representative
samples in China. This paper demonstrates FTIR combined with
chemometrics provides a useful tool for Halal authentication of similar ham sausages. However, if one wants to analyze ham sausages
with very different compositions and production procedure, it is recommended specic models be built with the samples of interest, because a model with very diverse training samples would have much
more complexity and degraded generalization performance.
Acknowledgements
This work was nancially supported by the National Public
Welfare Industry Projects of China (no. 201210010 and 201210092)
and Hangzhou Programs for Agricultural Science and Technology
Development (no. 20101032B28). Chen-Bo Cai is grateful to the
nancial aid of the Applied and Basic Research Project of Yunnan
Provincial Science and Technology Department (no. 2010CD087).
References
Aida, A. A., Che Man, Y. B., Raha, A. R., & Son, R. (2007). Detection of pig derivatives in
food products for Halal authentication by polymerase chain reactionrestriction
fragment length polymorphism. Journal of the Science of Food and Agriculture, 87,
569572.
Aida, A. A., Che Man, Y. B., Wong, C. M. V. L., Raha, A. R., & Son, R. (2005). Analysis of raw
meats and fats of pigs using polymerase chain reaction for Halal authentication.
Meat Science, 69, 4752.
Barker, M., & Rayens, W. (2003). Partial least squares for discrimination. Journal of
Chemometrics, 17, 166173.
Barnes, R. J., Dhanoa, M. S., & Lister, S. J. (1989). Standard normal variate transformation and detrending of near infrared diffuse reectance. Applied Spectroscopy, 43,
772777.
Che Man, Y. B., Gan, H. L., NorAini, I., Nazimah, S. A. H., & Tan, C. P. (2005). Detection of
lard adulteration in RBD palm olein using an electronic nose. Food Chemistry, 90,
829835.
Che Man, Y. B., & Mirghani, M. E. S. (2001). Detection of lard mixed with body fats of
chicken, lamb, and cow by Fourier transform infrared spectroscopy. Journal of the
American Oil Chemists' Society, 78, 753761.

Che Man, Y. B., Syahariza, Z. A., Mirghani, M. E. S., Jinap, S., & Bakar, J. (2005). Analysis of
potential lard adulteration in chocolate and chocolate products using Fourier
transform infrared spectroscopy. Food Chemistry, 90, 815819.
Coni, E., Pasquale, M. D., Cappolelli, P., & Bocca, A. (1994). Detection of animal fats in
butter by SC: A pilot study. Journal of the American Oil Chemists' Society, 71,
807810.
Farag, R. S., Abo-raya, S. H., Ahmed, F. A., Hewedi, F. M., & Khalifa, H. H. (1983). Fractional crystallization and gas chromatographic analysis of fatty acids as a means
of detecting butterfat adulteration. Journal of the American Oil Chemists' Society,
60, 16651669.
Forina, M., Armanino, C., Leardi, R., & Drava, G. (1991). A class modelling technique
based on potential functions. Journal of Chemometrics, 5, 435453.
Garip, S., Gozen, A. C., & Severcan, F. (2009). Use of Fourier transform infrared spectroscopy for rapid comparative analysis of Bacillus and Micrococcus isolates. Food
Chemistry, 113, 13011307.
Guillen, M. D., & Cabo, N. (1997). Characterization of edible oils and lard by Fourier
transform infrared spectroscopy. Relationships between composition and frequency of concrete bands in the ngerprint region. Journal of the American Oil Chemists'
Society, 74, 12811286.
Hargin, K. D. (1996). Authenticity issues in meat and meat products. Meat Science, 43,
277289.
Jaswir, I., Mirghani, M. E. S., Hassan, T. H., & Mohd Said, M. Z. (2003). Determination of
lard in mixtures of body fats of mutton and cow by Fourier transform-Infra red
(FTIR) spectroscopy. Journal of Oleo Science, 52, 633638.
Kokalj, M., Rihtari, M., & Kreft, S. (2011). Commonly applied smoothing of IR spectra
showed unappropriate for the identication of plant leaf samples. Chemometrics
and Intelligent Laboratory Systems, 108, 154161.
Kowalski, B. (1989). Sub-ambient differential scanning calorimetry of lard and lard
contaminated by tallow. International Journal of Food Science and Technology, 24,
415420.
Lockley, A. K., & Bardsley, R. G. (2000). DNA-based methods for food authentication.
Trends in Food Science and Technology, 11, 6777.
Marikkar, J. M. N., Ghazali, H. M., Che Man, Y. B., Peiris, T. S. G., & Lai, O. M. (2005). Distinguishing lard from other animal fats in admixtures of some vegetable oils using
liquid chromatographic data coupled with multivariate data analysis. Food Chemistry, 91, 514.
Meza-Mrquez, O. G., Gallardo-Velzquez, T., & Osorio-Revilla, G. (2010). Application
of mid-infrared spectroscopy with multivariate analysis and soft independent
modeling of class analogies (SIMCA) for the detection of adulterants in minced
beef. Meat Science, 86, 511519.
Murugaiah, C., Noor, Z. M., Mastakim, M., Bilung, L. M., Selamat, J., & Radu, S. (2009).
Meat species identication and Halal authentication analysis using mitochondrial
DNA. Meat Science, 83, 5761.
Rashood, K. A., Shaaban, R. R. A., Moety, E. M. A., & Rauf, A. (1995). Triacylglycerolsproling
by high performance liquid chromatography: A tool for detection of pork fat in
processed foods. Journal of Liquid Chromatography, 18, 2631.
Ripoche, A., & Guillard, A. S. (2001). Determination of fatty acid composition of pork fat
by Fourier transform infrared spectroscopy. Meat Science, 58, 299304.
Rohman, A., & Che Man, Y. B. (2010). FTIR spectroscopy combined with chemometrics
for analysis of lard in the mixtures with body fats of lamb, cow, and chicken. International Food Research Journal, 17, 519527.
Rohman, A., Sismindari, Erwanto, Y., & Che Man, Y. B. (2011). Analysis of pork adulteration in beef meatball using Fourier transform infrared (FTIR) spectroscopy. Meat
Science, 88, 9195.
Saeed, T., Ali, S. G., Rahman, H. A., & Sawaya, W. N. (1989). Detection of pork and lard as
adulterants in processed meat: Liquid chromatographic analysis of derivatized triglycerides. Journal of the Association of Ofcial Analytical Chemists, 72, 921925.
Savitzky, A., & Golay, M. J. E. (1964). Smoothing and differentiation of data by simplied least-squares procedures. Analytical Chemistry, 36, 16271639.
Snee, R. D. (1977). Validation of regression models: Methods and examples.
Technometrics, 19, 415428.
Suykens, J. A. K., & Vandewalle, J. (1999). Least squares support vector machine classiers. Neural Processing Letters, 9, 293300.
Syahariza, Z. A., Che Man, Y. B., Selamat, J., & Bakar, J. (2005). Detection of lard adulteration in cake formulation by Fourier transform infrared (FTIR) spectroscopy. Food
Chemistry, 92, 365371.
Xu, Q. -S., & Liang, Y. -Z. (2001). Monte Carlo cross validation. Chemometrics and Intelligent
Laboratory Systems, 56, 111.