American Journal of Anal yt ical Chemistry, 2011, 2, 135-141
doi:10.4236/ajac.2011.22015 Published Online May 2011 (
Copyright © 2011 SciRes. AJAC
Classification and Quantitative Analysis of Azithromycin
Tablets by Raman Spectroscopy and Chemometrics
Yan Li, Guorong Du, Wensheng Cai, Xueguang Shao
Research Center for Analytical Sciences, College of Chemistry, Nankai University, Tianjin, China
Received December 23, 2010; revised January 10, 2011; accepted January 21, 2011
Raman spectroscopy has been proven a noninvasive technique with high potential in pharmaceutical industry.
In this study, micro Raman technique and chemometric tools were used for identification of azithromycin
(AZM) tablets by different manufacturers and quantitative analysis of the active pharmaceutical ingredient
(API) in the samples. Support vector machine (SVM), Bayes classifier and K-nearest neighbour (KNN) were
employed for identification, partial least squares (PLS) regression was used for quantitative determination,
and interval partial least squares (iPLS) and Monte Carlo based uninformative variable elimination
(MC-UVE) methods were used to select informative variables for improving the models. The results show
that all the samples can be classified into groups by manufacturers with high accuracy, and the correlation
coefficient between the predicted API concentrations and reference values is as high as 0.96. Therefore, mi-
cro Raman spectroscopy coupled with chemometrics may be a fast and powerful tool for identification and
quantitative determination of pharmaceutical tablets.
Keywords: Azithromycin, Raman Spectroscopy, Pharmaceutical Tablets, Variable Selection, Partial Least
Squares (PLS)
1. Introduction
Azithromycin (AZM) is a second-generation macrolide
antibiotic. Owing to its superior antibacterial activity,
excellent pharmacokinetic properties and few gastroin-
testinal side effects compared to the first-generation
macrolide erythromycin, it has been widely used to treat
bacterial infections such as respiratory tract infections,
sexually transmitted diseases, skin and soft tissue infec-
tions [1]. Therefore, many methods have been developed
for detection of AZM, although they require time-con-
suming treatments prior to measurement.
Vibrational spectroscopic techniques, such as infrared
(IR), near infrared (NIR) and Raman spectroscopy, asso-
ciated with multivariate methods have been employed
widely in pharmaceutical industry due to their conven-
ience and efficiency [2,3]. Raman spectroscopy is known
as an excellent technique for the characterization of
pharmaceutical products, especially for the study of
solid-state medicaments, because it does not need ma-
nipulation or destruction of samples, can determine more
than one component at the same time, and avoids the
interference of water in the analyses. Raman technique
has thus been widely applied to the pharmaceutical in-
dustry and the characterization of tablets, such as ecstasy
[4], indomethacin [5] and ambroxol [6]. Applications of
Raman spectroscopy in tablet analysis has extensively
reported, such as solid-state evaluation, polymorphic
forms analysis, investigation of coating characteristics,
quantitative determination [7,8], and particularly in the
identification of pharmaceutical drugs [9] and counter-
feits [10-12]. Moreover, on-line analysis [13] and proc-
ess analysis [14] by using Raman spectroscopy were also
Application of chemometric techniques has made a
large improvement in the performance of Raman spec-
troscopy based methods for investigation of tablets. de
Veij [15] established an automated approach for detec-
tion of counterfeit artesunate tablets by utilizing Raman
spectroscopy and principal components analysis (PCA)
combined with hierarchical cluster analysis. Support
vector machines (SVMs) for classification of tablets of
different product families employing Raman technique
was studied by Roggo et al. [16]. Zhang et al. [17] ap-
plied Raman imaging to compare multivariate analysis
methods for classification and quantitative analysis of a
model pharmaceutical tablet. Partial least squares dis-
criminant analysis (PLS-DA) models based on Raman
spectra were applied in the counterfeits test by de Peinder
et al. [18]. Besides, multivariate calibration methods
have been applied to construct models for quantification
of the active pharmaceutical ingredients (APIs) in tablets
based on Raman spectra, such as partial least squares
(PLS) [19-22], principal component regression (PCR)
[23] and artificial neural networks (ANN) [24].
In this study, AZM tablets from different manufactur-
ers in China were studied by Raman spectroscopy and
chemometric methods for identification and quantifica-
tion. The classification models based on Bayes classifier
[25], K-nearest neighbor (KNN) [26,27], SVM [16,28,29]
and PLS-DA [18,30,31], were used to determine the
manufacturer and commonly used PLS regression was
adopted for quantification of the API. In addition, the
wavelength selection was investigated for improving the
performance of the models. The practicability was dem-
onstrated to use Raman spectroscopy for classification
and quantitative analysis of pharmaceutical tablets.
2. Experimental
2.1. Samples
31 AZM tablets of four different production enterprises
were analyzed. The samples from the four manufacturers
were referred to as A, B, C and D. Because the chemical
composition of the excipients may be different for dif-
ferent manufacturer, and variations may exist in different
batches of a product, samples from different batches
were selected. Concentrations of API in the tablets de-
termined by the reference method (HPLC) were provided
by National Institute for the Control of Pharmaceutical
and Biological Products (Beijing, China), ranging from
50% to 60% (w/w).
2.2. Measurement of Raman Spectra
Raman spectra of the samples were obtained with a
BRUKER SENTERRA Micro Raman spectrometer
(Bruker Optics Inc, Ettlingen, Germany) equipped with
charge coupled device (CCD) based on semiconductor
refrigeration technology and 785 nm laser. In the ex-
periments, a sample was placed on the microscope car-
rier, and located on the facula (50 1000 µm) of the
laser beam passing through the focusing systems. The
spectra were collected at a resolution of 8 cm1, over the
wavenumber range of 3500 - 70 cm1. Each spectrum is
composed of 1716 data points. The integration time was
10 s and 5 co-additions were performed at a laser power
of 100 mW on the sample. The OPUS (Bruker Optics Inc,
Ettlingen, Germany) software was used for data acquisi-
tion. The coating of the film-coated tablets was system-
atically removed with a scalpel before measurements.
2.3. Data Pretreatment
Figure 1(a) shows typical Raman spectra for the tablets
in four different categories. It can be seen that the spec-
tral intensity of the samples are quite different from each
other, which may be caused by crush size and dispersion
effects. Multiplicative scatter correction (MSC) is, there-
fore, used to correct the drifting baselines. Figure 1(b)
shows the spectra after MSC pretreatment. It is apparent
that the baseline was corrected in the pretreated spectra.
2.4. Classification and Quantification
Four classifiers were adopted for identification analysis,
i.e., Bayes, KNN, SVM and PLS-DA. All the classifiers
have been well studied for classification problems. Bayes
classifier is similar to linear discriminant analysis (LDA)
Wavenumber (cm
Raman intensity
1000 1500
3000 3500
Wavenumber (cm
Raman intensity
500 1000 1500 2000
3000 35
Figure 1. Raman spectra of the representative azithromycin
tablets in the four categories (a) and the pretreated spectra
with MSC (b).
Copyright © 2011 SciRes. AJAC
Y. LI ET AL.137
that is a linear method particularly suitable for high di-
mensional problems. KNN classifier is simply based on
the distance of the samples making it a common method
for classification problems. In this work, 2NN with
Euclidean distance was used. SVM is a nonlinear algo-
rithm by a kernel transformation mapping the samples
into a high dimensional hyperplane so that the examples
of the separate categories can be divided by a clear gap
as wide as possible. PLS-DA is the most commonly used
method for identification. Quantitative analysis was per-
formed by PLS regression, because it is the most com-
monly used multivariate calibration tool.
In the calculations, all the samples were divided into
calibration and validation set by a ratio of ca. 3:2 using
Kennard-Stone method [32]. Therefore, 21 samples were
used as calibration set to construct the classification and
quantitative models, and the other 10 samples were used
as validation set to evaluate the models. Each set covers
all the four categories of samples. The factor number for
PLS model was determined by the root mean square er-
ror of cross validation (RMSECV), which was obtained
by leave-one-out cross validation (LOO-CV). The per-
formance of classification model was evaluated by accu-
racy and PLS model was evaluated in terms of the corre-
lation coefficient (R) between the reference and predic-
tion concentration, the RMSECV and the root mean
square error of prediction (RMS E P) of the validation set.
2.5. Variable Selection Methods
Raman spectra are composed of 1716 variables in this
study as described above. However, not all the variables
have equivalent effect in the model. Some of them may
take important role in the model, but others may be un-
informative. Therefore, interval partial least squares
(iPLS) [33-35] and Monte Carlo based uninformative
variable elimination (MC-UVE) [36,37] were used for
variable selection for classification and quantitative
modeling, respectively. iPLS subdivides the data into
non-overlapping sections that each undergoes a separate
PLS modeling, and the useful variable ranges are deter-
mined by the value of RMSECV of the local models. The
stability was used in MC-UVE method to evaluate the
importance of each variable in the models, so that those
variables with larger stability are known as informative
and used in the modeling. It is worthy of noting that, in
the variable selection for quantitative models, the same
way as in the literatures [36,37] was used, for the classi-
fication models, however, the concentration vector was
replaced by the class number of the samples. The class
number for the samples of class A, B, C, and D were
assigned as 1, 2, 3 and 4, respectively.
3. Results and Discussions
3.1. Raman Spectra of AZM Tablets
Figure 2 shows a typical Raman spectrum of the samples
in category B, on which the vibrational bands assigned to
API and excipients are marked respectively. The tenta-
tive assignments are based on the comparison with the
published data in literatures [38-40]. The regions around
2934, 780 and 734 cm1 are related to the CH vibrations,
which are arisen from API and excipients in tablets. The
band at 1720 cm1 corresponds to CO double bond vibra-
tion, which is mainly caused by AZM in tablets. Mean-
while, the peaks between 1500 and 1000 cm1 should be
due to coupled vibrations in AZM. However, the Raman
bands at 968, 870 and 404 cm1 may be caused by PO
vibration, which is due to phosphate used as diluent in
pharmaceutical preparations. The Raman bands at 668
and 356 cm1 can be assigned to SiO vibration in talc,
which is commonly used in oral solid dosage formula-
tions as a lubricant and diluent. The band at 616 cm1
may be caused by the TiO2, which is used in pharmaceu-
tical formulations as a whitener [40]. The Raman band
around 474 cm1 can be assigned to the CC backbone
stretch in starch [40], which is the main excipient in
producing pharmaceutical formulations. Furthermore, the
bands between 400 and 200 cm1 and the strong band at
110 cm1 are probably caused by a combination of API
and excipients in tablets.
3.2. Classification
Table 1 summarized the classification results of the
samples in the validation set by Bayes, KNN, SVM and
PLS-DA, both the results obtained with full-spectrum
Wavenumber (cm
Raman intensity
3000 350
Figure 2. A representative spectrum of azithromycin tablets
in category B and the peaks assignment. Bands assigned to
API and excipients are marked by () and () respectively.
Copyright © 2011 SciRes. AJAC
Copyright © 2011 SciRes. AJAC
and selected wavenumbers by iPLS (partial-spectrum)
were listed. It is clear that, when the full-spectrum was
used, the accuracies are less than 80.00% for all the four
methods. The results indicate that not all the wavenum-
bers in the spectra are informative for the classification,
wavenumber selection is needed.
3.3. Quantification
At first, PLS model was built by using the 21 calibration
samples and full-spectrum data. The number of latent
variables used in this model was determined by leave-
one-out cross validation. The predicted results are sum-
marized in the first line of Table 2. It can be seen that
the correlation coefficient (R) between reference contents
and prediction values of the validation set is 0.9271, and
the recoveries are in the range of 95.87% - 102.65%.
For the purpose of improving the accuracy of classifi-
cation, iPLS was adopted to select informative variables.
Figure 3 shows the result obtained by iPLS and the dot
line shows the value of the full-spectrum model. It can be
seen that, most of local models are inferior to the full-
spectrum model and only the models of interval No. 2, 3,
4 and 17 surpass the full-spectrum model. Therefore, the
343 variables in the four intervals (two regions from 242
to 756 cm1 and from 2822 to 2990 cm1) were selected
to build the classification model. It is clear that the re-
gion from 242 to 756 cm1 is the information of inor-
ganic substances of the excipients in the samples, and the
band of 2822 - 2990 cm1 involves the vibration of CH,
which is related to the API in the samples.
Interval number
4 6
8 10 12 14
16 18
After variable selection, the classification accuracy of
PLS-DA reaches 100% as shown in Table 1. The results
show that Raman spectroscopy with the aid of chemom-
etrics may be a convenient tool for identification of
pharmaceutical tablets. However, the variable selection
has no significant improvement for the classification
accuracies of the other three methods. The reason may be
that the variables selection is done with iPLS, in which
the performance of each variable region is evaluated by
using the coefficients of the PLS-DA model. New meth-
ods should be developed for improving the results of the
Figure 3. RMSECVs obtained by iPLS for the 20 local mod-
els. The dot line indicates the RMSECV of the full-spectrum
Table 1. Classification results for the samples in the validation set.
Classification accuracy (%)
Classifier Classification criteria
Full-spectrum Partial-spectrum (Na)
Bayes Diaglinearb 70.00 70.00 (343)
KNN Euclidean distancec 80.00 60.00 (343)
SVM Linear kerneld 80.00 60.00 (343)
PLS-DA 80.00 100.00 (343)
aN is the number of wavenumbers used in the models; bDiaglinear is a type of discriminant function that calculates the density of each group
with a diagonal covariance matrix; cEuclidean distance measures the distance of two points using the Pythagorean formula; dLinear kernel
produces a linear subspace of a homogeneous system with linear equations.
Table 2. Results of the PLS models for quantification.
Spectral range Latent variable number RMSEP R Recoveries (%)
Full-spectrum 5 0.0120 0.9271 95.87 - 102.65
Partial-spectrum (N = 135) 5 0.0080 0.9634 98.29 - 102.65
Copyright © 2011 SciRes. AJAC
Although the results obtained with full-spectrum data
are acceptable, variable selection was performed by us-
ing the MC-UVE method for the purpose of simplifying
and further improving the PLS model. Figure 4 shows
the distribution of the stability of each variable in the
wavenumber 70 - 3500 cm1 by MC-UVE method. Clear-
ly, there are lots of the variables whose stability absolute
values are very small. This indicates that most of the
variables are uninformative and can be eliminated, and
quantification model with those variables whose stability
absolute values are comparatively large should be better.
Therefore, the variation of the RMSECV with the number
of the retained variables was investigated. Figure 5
shows the RMSECV obtained with different number of
variables selected according to the stability values. It can
be seen that, when N is 135, the lowest value of RMSECV
is obtained. In order to build a parsimonious PLS model,
135 variables are used, and the threshold determined by
Wavenumber (cm
Raman intensity
3000 3500
Figure 4. Stability of the variables obtained by MC-UVE.
The dot lines indicate the threshold.
Number of variables
0 200
400 600 800 1000
Figure 5. Variation of RMSECV with the number of selected
the number is marked with dot lines in Figure 4. It can
be found that the selected variables are concentrated on
the bands below 1500 cm1, four broad bands around 712,
1168, 1206 and 1468 cm1, some narrow wavelength
intervals around 190, 316, 458, 602, 618, 736, 974, 1002
and 1056 cm1, and several wavenumbers around 2610
and 2768 cm1. The bands in these regions are mainly
related to the API in tablets.
The second line of Table 2 shows the predicted results
of the validation set with the simplified PLS model. The
correlation coefficient and the recoveries are improved to
0.9634 and 98.29% - 102.65%, respectively. Compared
with the full-spectrum model, the results are further im-
proved. Table 3 demonstrates the prediction error for
each sample in the validation set. All these values can be
found to be within 3% (the maximum is 2.65% corre-
sponding to the maximal recovery of 102.65%). Such
results clearly show that Raman spectroscopy with the
aid of chemometrics is a convenient and precise tool for
quantification of API in tablets.
4. Conclusions
The flexibility of identification and quantitative deter-
mination of AZM tablets by using micro Raman spec-
troscopy is investigated. The accuracy of a PLS-DA
model with full-spectral data is as high as 80.00%, and
the model with the variables selected by iPLS can reach
100%. For quantitative determination, the full-spectrum
PLS model produces prediction accuracy of 95.87% -
102.65%, and the partial-spectrum PLS model with the
variables selected by MC-UVE can further improve the
prediction accuracy to 98.29% - 102.65%. Therefore, this
study demonstrates that Raman spectroscopy coupled
with appropriate chemometric methods provides a con-
venient way for fast identification and quantification of
tablets. The method may take an important role for the
quality control of release pharmaceutical tablets.
Table 3. Predicted values and the relative errors for the
samples of the validation set.
ClassSample No.Reference
(%, w/w)
PLS model
(%, w/w)
Relative err
6 57.10 58.61 2.65
C 8 57.10 57.89 1.38
D 12 59.82 59.66 –0.26
17 51.93 51.71 –0.43
18 51.93 52.84 1.76
19 51.42 52.02 1.18
22 52.95 53.52 1.07
26 53.72 52.80 –1.71
29 53.97 53.13 –1.55 B
30 54.68 55.34 1.21
5. Acknowledgements
The authors thank Dr. Feng, National Institute for the
Control of Pharmaceutical and Biological Products, for
providing the samples and the related information, and
Bruker Optics-Beijing for instrumental support. This
study is supported by National Natural Science Founda-
tion of China (No. 20835002).
6. References
[1] K. A. Shaikh, S. D. Patil and A. B. Devkhile, “Develop-
ment and Validation of a Reversed-Phase HPLC Method
for Simultaneous Estimation of Ambroxol Hydrochloride
and Azithromycin in Tablet Dosage Form,” Journal of
Pharmaceutical and Biomedical Analysis, Vol. 48, No. 5,
2008, pp. 1481-1484. doi:10.1016/j.jpba.2008.09.031
[2] C. Gendrin, Y. Roggo and C. Collet, “Pharmaceutical
Applications of Vibrational Chemical Imaging and Che-
mometrics: A Review,” Journal of Pharmaceutical and
Biomedical Analysis, Vol. 48, No. 3-4, 2008, pp. 533-553.
[3] S. H. F. Scafi and C. Pasquini, “Identification of Coun-
terfeit Drugs Using Near-Infrared Spectroscopy,” Analyst,
Vol. 126, No. 12, 2001, pp. 2218-2224.
[4] S. E. J. Bell, D. T. Burns, A. C. Dennis, L. J. Matchett
and J. S. Speers, “Composition Profiling of Seized Ec-
stasy Tablets by Raman Spectroscopy,” Analyst, Vol. 125,
No. 10, 2000, pp. 1811-1815. doi:10.1039/b005662f
[5] T. Okumura and M. Otsuka, “Evaluation of the Micro-
crystallinity of a Drug Substance, Indomethacin, in a
Pharmaceutical Model Tablet by Chemometric FT-Ram-
an Spectroscopy,” Pharmaceutical Research, Vol. 22, No.
8, 2005, pp. 1350-1357.
[6] M. S. Hwang, S. Cho, H. Chung and Y. A. Woo, “Non-
destructive Determination of the Ambroxol Content in
Tablets by Raman Spectroscopy,” Journal of Pharma-
ceutical and Biomedical Analysis, Vol. 38, No. 2, 2005,
pp. 210-215. doi:10.1016/j.jpba.2004.12.031
[7] M. Dyrby, S. B. Engelsen, L. Norgaard, M. Bruhn and L.
Lundsberg-Nielsen, “Chemometric Quantitation of the
Active Substance (Containing C N) in a Pharmaceutical
Tablet Using Near-Infrared (NIR) Transmittance and NIR
FT-Raman Spectra,” Applied Spectroscopy, Vol. 56, No.
5, 2002, pp. 579-585. doi:10.1366/0003702021955358
[8] S. Sasic, “An In-Depth Analysis of Raman and Near-In-
frared Chemical Images of Common Pharmaceutical
Tablets,” Applied Spectroscopy, Vol. 61, No. 3, 2007, pp.
239-250. doi:10.1366/000370207780220769
[9] S. E. J. Bell, L. A. Fido, N. M. S. Sirimuthu, S. J. Speers,
K. L. Peters and S. H. Cosbey, “Screening Tablets for
DOB Using Surface-Enhanced Raman Spectroscopy,”
Journal of Forensic Sciences, Vol. 52, No. 5, 2007, pp.
1063-1067. doi:10.1111/j.1556-4029.2007.00515.x
[10] C. Ricci, C. Eliasson, N. A. Macleod, P. N. Newton, P.
Matousek and S. G. Kazarian, “Characterization of Ge-
nuine and Fake Artesunate Anti-Malarial Tablets Using
Fourier Transform Infrared Imaging and Spatially Offset
Raman Spectroscopy Through Blister Packs,” Analytical
and Bioanalytical Chemistry, Vol. 389, No. 5, 2007, pp.
1525- 1532. doi:10.1007/s00216-007-1543-1
[11] C. Ricci, L. Nyadong, F. Yang, F. M. Fernandez, C. D.
Brown, P. N. Newton and S. G. Kazarian, “Assessment of
Hand-Held Raman Instrumentation for in situ Screening
for Potentially Counterfeit Artesunate Antimalarial Tab-
lets by FT-Raman Spectroscopy and Direct Ionization
mass Spectrometry,” Analytica Chimica Acta, Vol. 623,
No. 2, 2008, pp. 178-186. doi:10.1016/j.aca.2008.06.007
[12] C. Eliasson and P. Matousek, “Noninvasive Authentica-
tion of Pharmaceutical Products Through Packaging Us-
ing Spatially Offset Raman Spectroscopy,” Analytical
Chemistry, Vol. 79, No. 4, 2007, pp. 1696-1701.
[13] P. Matousek and A. W. Parker, “Bulk Raman Analysis of
Pharmaceutical Tablets,” Applied Spectroscopy, Vol. 60,
No. 12, 2006, pp. 1353-1357.
[14] R. B. Shah, M. A. Tawakkul and M. A. Khan, “Process
Analytical Technology: Chemometric Analysis of Raman
And Near Infrared Spectroscopic Data for Predicting
Physical Properties of Extended Release Matrix Tablets,”
Journal of Pharmaceutical Sciences, Vol. 96, No. 5, 2007,
pp. 1356-1365. doi:10.1002/jps.20931
[15] M. de Veij, P. Vandenabeele, K. A. Hall, F. M. Fernan-
dez, M. D. Green, N. J. White, A. M. Dondorp, P. N.
Newton and L. Moens, “Fast Detection and Identification
of Counterfeit Antimalarial Tablets by Raman Spectros-
copy,” Journal of Raman Spectroscopy, Vol. 38, No. 2,
2007, pp. 181-187. doi:10.1002/jrs.1621
[16] Y. Roggo, K. Degardin and P. Margot, “Identification of
Pharmaceutical Tablets by Raman Spectroscopy and Che-
mometrics,” Talanta, Vol. 81, No. 3, 2010, pp. 988-995.
[17] L. Zhang, M. J. Henson and S. S. Sekulic, “Multivariate
Data Analysis for Raman Imaging of a Model Pharma-
ceutical Tablet,” Analytica Chimica Acta, Vol. 545, No. 2,
2005, pp. 262-278. doi:10.1016/j.aca.2005.04.080
[18] P. de Peinder, M. J. Vredenbregt, T. Visser and D. de
Kaste, “Detection of Lipitor (R) Counterfeits: A Com-
parison of NIR And Raman Spectroscopy in Combination
with Chemometrics,” Journal of Pharmaceutical and
Biomedical Analysis, Vol. 47, No. 4-5, 2008, pp. 688-694.
[19] R. Szostak and S. Mazurek, “FT-Raman Quantitative
Determination of Ambroxol in Tablets,” Journal of Mo-
lecular Structure, Vol. 704, No. 1-3, 2004, pp. 229-233.
[20] J. Johansson, A. Sparen, O. Svensson, S. Folestad and M.
Claybourn, “Quantitative Transmission Raman Spectros-
copy of Pharmaceutical Tablets and Capsule,” Applied
Spectroscopy, Vol. 61, No. 11, 2007, pp. 1211-1218.
[21] R. Szostak and S. Mazurek, “Quantitative Determination
Copyright © 2011 SciRes. AJAC
Copyright © 2011 SciRes. AJAC
of Acetylsalicylic Acid and Acetaminophen in Tablets by
FT-Raman Spectroscopy,” Analyst, Vol. 127, No. 1, 2002,
pp. 144-148. doi:10.1039/b108240j
[22] S. Mazurek and R. Szostak, “Quantification of Atorvas-
tatin Calcium in Tablets by FT-Raman Spectroscopy,”
Journal of Pharmaceutical and Biomedical Analysis, Vol.
49, No. 1, 2009, pp. 168-172.
[23] S. Mazurek and R. Szostak, “Quantitative Determination
of Captopril and Prednisolone in Tablets by FT-Raman
Spectroscopy,” Journal of Pharmaceutical and Biomedi-
cal Analysis, Vol. 40, No. 5, 2006, pp. 1225-1230.
[24] S. Mazurek and R. Szostak, “Quantitative Determination
of Diclofenac Sodium in Solid Dosage Forms by
FT-Raman Spectroscopy,” Journal of Pharmaceutical and
Biomedical Analysis, Vol. 48, No. 3, 2008, pp. 814-821.
[25] H. Sun, “A Naive Bayes Classifier for Prediction of Mul-
tidrug Resistance Reversal Activity on the Basis of Atom
typing,” Journal of Medicinal Chemistry, Vol. 48, No. 12,
2005, pp. 4031-4039. doi:10.1021/jm050180t
[26] K. C. Chou and H. B. Shen, “Predicting Eukaryotic Pro-
tein Subcellular Location by Fusing Optimized Evi-
dence-Theoretic K-Nearest Neighbor Classifiers,” Journal
of Proteome Research, Vol. 5, No. 8, 2006, pp. 1888-1897.
[27] U. Schmid, P. Rosch, M. Krause, M. Harz, J. Popp and K.
Baumann, “Gaussian Mixture Discriminant Analysis for
the Single-Cell Differentiation of Bacteria Using Mi-
cro-Raman Spectroscopy,” Chemometrics and Intelligent
Laboratory Systems, Vol. 96, No. 2, 2009, pp. 159-171.
[28] R. Karchin, K. Karplus and D. Haussler, “Classifying G-
protein Coupled Receptors with Support Vector Mac-
hines,” Bioinformatics, Vol. 18, No. 1, 2002, pp. 147-159.
[29] S. Zomer, R. G. Brereton, J. F. Carter and C. Eckers,
“Support Vector Machines for the Discrimination of
Analytical Chemical Data: Application to the Determina-
tion of Tablet Production by Pyrolysis-Gas Chromatog-
raphy-Mass Spectrometry,” Analyst, Vol. 129, No. 2,
2004, pp. 175-181. doi:10.1039/b312982a
[30] C. D. Natale, A. Macagnano, E. Martinelli, R. Paolesse,
G. D'Arcangelo, C. Roscioni, A. Finazzi-Agro and A.
D'Amico, “Lung Cancer Identification by the Analysis of
Breath by Means of an Array of Non-Selective Gas Sen-
sors,” Biosensors and Bioelectronics, Vol. 18, No. 10, 2003,
pp. 1209-1218. doi:10.1016/S0956-5663(03)00086-1
[31] S. Bijlsma, L. Bobeldijk, E. R. Verheij, R. Ramaker, S.
Kochhar, I. A. Macdonald, B. van Ommen and A. K.
Smilde, “Large-Scale Human Metabolomics Studies: A
Strategy For Data (Pre-) Processing And Validation,”
Analytical Chemistry, Vol. 78, No. 2, 2006, pp. 567-574.
[32] R. W. Kennard and L. A. Stone, “Computer Aided De-
sign of Experiments,” Technometrics, Vol. 11, 1969, pp.
137-148. doi:10.2307/1266770
[33] L. Norgaard, A. Saudland, J. Wagner, J. P. Nielsen, L.
Munck and S. B. Engelsen, “Interval Partial Least-
Squares Regression (iPLS): A Comparative Chemometric
Study with an Example from Near-Infrared Spectros-
copy,” Applied Spectroscopy, Vol. 54, No. 3, 2000, pp.
413-419. doi:10.1366/0003702001949500
[34] C. Abrahamsson, J. Johansson, A. Sparen and F. Lindg-
ren, “Comparison of Different Variable Selection Meth-
ods Conducted on NIR Transmission Measurements on
Intact Tablets,” Chemometrics and Intelligent Laboratory
Systems, Vol. 69, No. 1, 2003, pp. 3-12.
[35] X. B. Zou, J. W. Zhao, M. J. W. Povey, M. Holmes and
H. P. Mao, “Variables Selection Methods in Near-In-
frared Spectroscopy,” Analytica Chimca Acta, Vol. 667,
No. 1, 2010, pp. 14-32.doi:10.1016/j.aca.2010.03.048
[36] W. S. Cai, Y. K. Li and X. G. Shao, “A Variable Selec-
tion Method Based on Uninformative Variable Elimina-
tion for Multivariate Calibration of Near-Infrared Spec-
tra,” Chemometrics and Intelligent Laboratory Systems,
Vol. 90, No. 2, 2008, pp. 188-194.
[37] H. Xu, Z. C. Liu, W. S. Cai and X. G. Shao, “A Wave-
Length Selection Method Based on Randomization Test
for Near-Infrared Spectral Analysis,” Chemometrics and
Intelligent Laboratory Systems, Vol. 97, No. 2, 2009, pp.
189-193. doi:10.1016/j.chemolab.2009.04.006
[38] P. Vandenabeele, B. Wehling, L. Moens, H. Edwards, M.
De Reu and G. Van Hooydonk, “Analysis with Micro-R-
aman Spectroscopy of Natural Organic Binding Media
and Varnishes Used in Art,” Analytica Chimica Acta, Vol.
407, No. 1-2, 2000, pp. 261-274.
[39] J. De Gelder, K. De Gussem, P. Vandenabeele and L.
Moens, “Reference Database of Raman Spectra of Bio-
logical Molecules,” Journal of Raman Spectroscopy, Vol.
38, No. 9, 2007, pp. 1133-1147. doi:10.1002/jrs.1734
[40] M. de Veij, P. Vandenabeele, T. De Beer, J. P. Remon
and L. Moens, “Reference Database of Raman Spectra of
Pharmaceutical Excipients,” Journal of Raman Spectros-
copy, Vol. 40, No. 3, 2009, pp. 297-307.