Classi
fi
cation of cereal bars using near infrared spectroscopy and linear
discriminant analysis
Anna Luiza Bizerra Brito
a, Lívia Rodrigues Brito
b, Fernanda Araújo Honorato
c,
Márcio José Coelho Pontes
a, Liliana Fátima Bezerra Lira Pontes
a,⁎
a
Universidade Federal da Paraíba, Departamento de Química, João Pessoa, PB, Brazil
b
Universidade Federal de Pernambuco, Departamento de Química Fundamental, Recife PE, Brazil
c
Universidade Federal de Pernambuco, Departamento de Engenharia Química, Recife PE, Brazil
a b s t r a c t
a r t i c l e i n f o
Article history: Received 5 November 2012 Accepted 9 February 2013 Keywords: Cereal barNear infrared spectroscopy Linear discriminant analysis Successive projection algorithm Genetic algorithm
Stepwise formulation
This work proposes an analytical method for cereal bar classification based on the use of near infrared spec-troscopy (NIRS) and supervised pattern recognition techniques. Linear discriminant analysis (LDA) is employed to build a classification model on the basis of a reduced subset of variables (wavenumbers). For the purpose of variable selection, three techniques are considered, namely successive projection algorithm (SPA), Genetic Algorithm (GA), and stepwise (SW) formulation. The methodology is validated in a case study involving the classification of 121 cereal bar samples into three different types (conventional, diet and light). The results show that the LDA/GA model is superior to the LDA/SPA and LDA/SW models with re-spect to classification accuracy in an independent prediction set. Some advantages of the proposed method are speed, that the analytical measurement is performed quickly (one minute or less per sample), no re-agents, low sample consumption and minimum sample preparation demands. In view of the results obtained in this study the proposed method may be considered valid for use in cereal bar classification.
© 2013 Elsevier Ltd.
1. Introduction
Consumption of foods which have low calorie content has increased considerably (Bingol, Zhang, Pan, & McHugh, 2012; Carrillo, Varela, & Fiszman, 2012; Gerend, 2009; Kocken, Buijs, & Snel, 2006). With so many options on the market, however, the consumerfinds it confusing to select the product they want among foods with labels such as“ natu-ral”,“light”,“diet”,“organic”, and“functional”, among others.
Light and diet foods can been found in most supermarkets and labeled as products with low fat, salt, protein, carbohydrates or sugar contents. According to the National Health Surveillance Agency (ANVISA) (Brazilian Agency of Sanitary Surveillance, 2012) these foods are specially formulated with modifications in nutrient content, suitable for use in different diets or for people with specific physio-logical and metabolic conditions.
The term“light”can be used for a food when the quantity of calories or nutrients is at least 25% less than the conventional product. According to ANVISA,“diet”can be applied to foods with an absence of sucrose/glucose or foods indicated for diets with restriction of some nutrients such as fat, carbohydrate, protein and sodium (Brazilian Agency of Sanitary Surveillance, 2012).
In order to complement regular food, products such as cereal bars are widely consumed as a fast snack with low caloric value. These foods were introduced nearly a decade ago, when consumers became particularly interested in health and diet, since the cereal bars have nutrients such asfibers, vitamins and minerals. Today, cereal bars are consumed worldwide, including by people on diets, or those with health problems or just as a quick snack (Farinazzi-Machado, Barbalho, Oshiiwa, Goulart, & Pessan Junior, 2012; Lobato et al., 2012; Villavicencio, Araújo, Fanaro, Rela, & Mancini-Filho, 2007; Zaveri & Drummond, 2009). Because they are easy to carry and are available on the market in different types, brands, flavors and nutritional compositions, these foods are very well adapted in the day-to-day lives of modern people.
For the consumer, the choice of the product is associated with its appearance, the description on the package and the nutritional infor-mation given. These parameters are not always effective, however, in guaranteeing a safe choice of the desired food. Incorrect information on the package or food label, such as the absence or presence of sugar or carbohydrates, can lead to incorrect choices by diabetics, for example, causing hyperglycemia and cardiovascular problems (Scott et al., 2011).
The quality control of cereal bars is performed through physical and chemical tests, such as protein content using the method de-scribed by Kjeldahl (AOAC, 1997), moisture determination (Adolfo Lutz Institute, 1985), lipid content (AOAC, 1997) as well as sensory analysis. These reference methods have the drawback of being ⁎ Corresponding author at: Universidade Federal da Paraíba, Departamento de Química,
Laboratório de Combustíveis e Materiais (LACOM), CEP 58051-970, João Pessoa, PB, Brazil. Tel./fax: +55 83 3216 7441.
E-mail address:[email protected](L.F.B.L. Pontes).
0963-9969 © 2013 Elsevier Ltd.
http://dx.doi.org/10.1016/j.foodres.2013.02.014
Contents lists available atSciVerse ScienceDirect
Food Research International
j o u r n a l h o m e p a g e : w w w . e l s e v i e r . c o m / l o c a t e / f o o d r e sOpen access under the Elsevier OA license.
destructive, tedious, time-consuming, using huge amounts of toxic chemical reagents, expensive and, in some cases, subjective.
Infrared spectroscopy (IR) is a non-destructive alternative ana-lytical technique which allows reliable, direct and fast determination of different properties at the same time without sample pre-treat-ment (Pasquini, 2003). Several papers in the literature report the use of near infrared (NIR) spectroscopy to monitor the quality of foods (Riovanto, Cynkar, Berzaghi, & Cozzolino, 2011; Sinelli, Cerretani, Di Egidio, Bendini, & Casiraghi, 2010).
In chemometrics data analysis, pattern recognition methods are a powerful tool in context of food quality assessment and food composi-tion analysis (Berrueta, Alonso-Salces, & Héberger, 2007; Cen & He, 2007). These methods have been successfully applied to classify a num-ber of foods, including yogurt (Cruz et al., 2013), beer (Granato, Branco, Faria, & Cruz, 2011), coffee (Souto et al., 2010), and coconut oil (Rohman & Che Man, 2011), among others (Berrueta et al., 2007; Cen & He, 2007).
Supervised pattern recognition methods essentially differ in the way they define classification rules. Basically, they can be divided into dis-criminating and class-modelling methods (Roggo, Duponchel, & Huvenne, 2003). Linear discriminant analysis (LDA) and partial least square-discriminant analysis (PLS-DA) are examples of discriminating techniques, whereas the soft independent modelling of class analogy (SIMCA) is a class-modelling method. The modeling strategies among these methods are substantially different. LDA and PLS-DA focus on the dissimilarity between classes and the samples must be classified in a particular training classe (Roggo et al., 2003; Vaid, Burl, & Lewis, 2001). The SIMCA method, however, considers each class separately and performs outlier tests to decide whether a new object belongs to a certain class, to all classes or does not belong to any class. The SIMCA method is frequently used in data sets with high dimensionality (Brereton, 2009) such as spectroscopic data. However, the testing pro-cedure adopted by SIMCA has the disadvantage that one has to set a confidence level,α. If the data are normally distributed,α% (e.g. 5%) of objects belonging to the class will be considered as not belonging to it. This misclassification problem can be avoided when the LDA method is employed. The LDA method employs linear decision boundaries, which are defined in order to maximize the ratio of between-class to within-class dispersion (Fisher, 1936). Its has been successfully applied to a number of classification problems (Gambarra Neto et al., 2009; Gori, Maggio, Cerretani, Nocetti, & Caboni, 2012; Riovanto et al., 2011; Sinelli et al., 2010; Souto et al., 2010). When compared with SIMCA and PLS-DA, the LDA method has the disadvantage that the number of training samples must be larger than the number of variables included in the LDA model. Therefore, procedures based on the selection of each variable are required for the classification of spectral data. The suc-cessive projections algorithm (SPA) (Pontes et al., 2005; Pontes, Pereira, Pimentel, Vasconcelos, & Silva; Silva, Pontes, Pimentel, & Pontes, 2012; Silva, Borba, et al., 2012), genetic algorithm (GA) (Pontes et al., 2005) and stepwise (SW) formulation (Caneca et al., 2006) methods have been adopted for this purpose in different classification problems.
In general, the literature reports only on works involving the characterization of cereal bars as well as new food products being developed for the market by conventional methods (Egert et al., 2012; Fonseca, Santo, Souza, & Pereira, 2011; Heenan et al., 2012; Santos et al., 2011). Explicit studies regarding the classification of ce-real bars with respect to type (conventional, diet and light) using NIR spectrometry and multivariate classification with wavenumber se-lection have not been found in the specialized literature. Thus, devel-opment of rapid and accurate methodologies is important for the economy and public health in order to identify non-conformity in ce-real bar samples.
In the present work, an analytical method to classify cereal bar sam-ples according to type (diet, light, conventional) using NIR spectroscopy and pattern recognition technique is proposed. For this, LDA is employed to build a classification model on the basis of a reduced
subset of spectral variables selected using three different techniques: successive projection algorithm (SPA) (Moreira, Pontes, Galvão, & Araújo, 2009; Pontes et al., 2005), the genetic algorithm (GA) (Pontes et al., 2005), and a stepwise (SW) formulation (Caneca et al., 2006). The results obtained by these three methods (LDA/SPA, LDA/GA, and LDA/SW) have been assessed in terms of classification results in a set of samples not used in the model-building process (prediction samples).
2. Material and methods 2.1. Samples
One hundred and twenty-one samples of three different types of cereal bars were analyzed: Diet (35); Conventional (44) and Light (42). All samples were crushed and sieved with 20 mesh particle size. 2.2. NIR spectra measurements
A Spectrum 400 FT–IR/FT–NIR spectrophotometer (Perkin Elmer) equipped with accessory reflectance (NIRA) was employed to obtain NIR spectra in the range of 10,000 a 4000 cm−1. All spectra were recorded with an average of 16 scans, and a spectral resolution of 8 cm−1. The background spectra were obtained using the Spectralon standard. Temperature was controlled at 23 ±1 °C throughout the spectral acquisition process.
2.3. Chemometric procedure
Raw spectra and some pre-processing strategies such as baseline correction, standard normal variate (SNV) (Barnes, Dhanoa, & Lister, 1989), smoothing and the Savitzky–Golayfirst derivative (Savitzky & Golay, 1964), second-order polynomial (7, 11 and 15 window points), were evaluated in terms of overall classification errors. Detection and elimination of outliers were carried out using score, residual and lever-age plots. For each type, the Kennard–Stone (KS) algorithm (Kennard & Stone, 1969) was applied in order to divide the samples into training, validation and prediction subsets.Table 1presents the number of sam-ples in each set. The validation method employed in this study is based on the test set, thereby the training and validation samples were used in the modelling procedures (including variable selection for LDA) where-as the prediction samples were only used in thefinal evaluation of the classification models. All spectral data were mean-centered before modeling procedures.
An exploratory analysis was initially performed in order to observe the existence of natural groupings. For this purpose, the principal com-ponent analysis (PCA) (Souza et al., 2011) was applied to the overall spectral data set.
The present work adopts the stepwise (SW) formulation developed byCaneca et al. (2006). In order to decide which wavenumbers were to be discarded, seven threshold values (0.30, 0.50, 0.70, 0.80, 0.90, 0.95, and 0.99) of multiple correlation coefficients were tested in the LDA/ SW model. The best threshold was selected on the basis of the classifi -cation errors in the validation set.
The SPA algorithm utilized in this paper (Pontes et al., 2005) em-ploys, as the cost function, the average risk G of misclassification by Table 1
Number of training, validation and prediction samples in each cereal bar type.
Type Sets
Training Validation Prediction
Diet 17 09 09
Conventional 22 11 11
Light 20 11 11
LDA which is calculated in the validation set, as presented in Eq.(1). The subset of variables which presented the lowest value of G was then selected by the LDA/SPA algorithm.
G¼ð1=kvÞX Kv
k¼1
gk; ð1Þ
wheregkis the risk of misclassification of thekthvalidation objectxk, k= 1,…,Kv. More details about the LDA/SPA can be found inPontes et al. (2005)andMoreira et al. (2009).
The GA algorithm formulation presented inPontes et al. (2005)was used in this study, where afitness value was defined for each chromo-some as the inverse of the average risk G of misclassification by LDA, as defined in Eq.(1). The GA routine was carried out over 100 genera-tions with 200 chromosomes in each generation. Mutation and cross-over probabilities were set to 10% and 60%, respectively. Because of the stochastic nature of the LDA/GA algorithm, the routine was repeated ten times (starting from different random initial populations) and the best solution found during these 10 executions was retained.
Spectral acquisition was obtained with Spectrum Version 6.3.4. Data preprocessing and PCA were carried out using Unscrambler® X.1 (CAMO ASA, Norway). The KS, LDA/SW, LDA/SPA and LDA/GA algo-rithms were coded in Matlab version 6.5 (Mathworks, USA).
3. Results and discussion 3.1. NIR spectra
Fig. 1 presents the raw NIR spectra of 121 cereal bar samples recorded in the range of 10,000–4000 cm−1.
The spectra show the characteristics bands of carbohydrates such as starch and sugars at 4760, 6711, 6740 cm−1; proteins at 4878–4854, 6667–6536 cm−1; fats at 4200, 4830 cm−1 and moisture at 5135, 8310 cm−1(Workman, 1996; Workman & Weyer, 2008).
3.2. Principal component analysis
Fig. 2presents the PC2× PC1 score plot resulting from the applica-tion of PCA to the raw NIR spectra. As can be seen, there is a substantial dispersion and overlapping of the cereal bar types. Therefore, spectral variable selection procedures are required in order to obtain a better discrimination among the three cereal bar types studied. PC1 and PC2 explain 94% and 2% of the data variance, respectively.
3.3. LDA classification
The best validation results were achieved with the raw spectra (without preprocessing). The variable numbers selected by SW, SPA and GA were 4, 12 and 17, respectively. Among the LDA models, the worst validation results in terms of correct classification rate (CCR) were obtained with SW, which only correctly classified 21 of the 31 validation samples. This outcome corresponds to a CCR of 68%. LDA/GA, however, provided better results in the validation set for all three cereal bar types, where only one light sample was incor-rectly classified as a conventional type sample. The CCR in validation set obtained by LDA/GA and LDA/SPA models were 97%, and 90%, re-spectively. Possibly, the best results were achieved by LDA/GA be-cause it selects wavenumbers around 8040 cm−1and 8500 cm−1, which can be attributed to second overtone of C–H stretching of su-crose (Filho, 2009; Workman, 1996; Workman & Weyer, 2008). It is worth noting that these variables were not selected by LDA/SW and LDA/SPA algorithms. The wavenumbers selected by SW, SPA and GA algorithms are presented inFig. 3.
It can be observed that the few variables selected by SW (Fig. 3a) are located in the region between 4300 and 5800 cm−1which can be attributed to presence of a second overtone C–H band associated with fat and combination C–H stretching. The isolated wavenumber selected by SW (10,000 cm−1) can be attributed to the second over-tone N–H stretch associated with protein. These variables were among those selected by SPA and GA algorithm (Fig. 3b and c). Additionally, peaks around 8040 and 8500 cm−1(second overtone
Fig. 1.Raw NIR spectra of 121 cereal bar samples.
Fig. 2.PC2×PC1 score plot for the overall set of 121 cereal bar samples ( :conventional; b: diet and :light).
of C–H stretching of sucrose) and 9440 cm−1(second overtone of O–H stretch) were selected by GA. Furthermore, the SPA and GA algorithms selected variables around 6200–6900 cm−1which may be attributed to the O–H polymeric of carbohydrates (Filho, 2009; Workman, 1996; Workman & Weyer, 2008).
The LDA models obtained with the variables selected by SW, SPA and GA algorithms were applied to the classification of the prediction set.Table 2summarizes the results of the LDA/SW, LDA/SPA and LDA/GA models in terms of correct classifications (predicted type index equal to correct type index) and incorrect classifications (predicted type index different from correct type index) for the prediction set. The CCR for each model is also presented inTable 2.
The best results of LDA models were achieved with GA which cor-rectly classified 28 of the 31 prediction samples. This classification per-formance was not achieved by LDA/SW, which had a correct prediction rate of only 61%. As can be seen inTable 2, the most frequent errors oc-curred between the light and conventional types. In fact, these foods have the same nutrients, but sometimes in different concentrations.
The classification performance of LDA models built with wave-numbers selected is further demonstrated byFig. 4a–c, which illustrate the scores of thefirst two discriminant functions (DF2× DF1) for the overall data set.
The best discrimination was found when LDA was applied to wavenumbers selected by GA (Fig. 4c) and SPA (Fig. 4b). In these two cases, the diet samples were separated from the light and con-ventional samples along DF1 direction. In addition, DF2 distinguished conventional samples from the diet and light types.
It is worth noting that the methodology employed was based solely on spectroscopic measurements and pattern recognition techniques. This indicates that laborious procedures for chemical characterization of the cereal bars are not required. However, these classification models cannot be applied to the analysis of other foods different from those in-cluded in this study. For different food samples, new classification models should be built and validated.
4. Conclusions
This paper proposed a methodology for cereal bar classification employing NIR spectroscopy and linear discriminant analysis coupled with the variable selection algorithms. In a case study involving three different cereal bar types (conventional, diet and light), the results indi-cate that wavenumber selection is key to an accurate discrimination of the samples. More specifically, the best classification results were achieved with the LDA/GA model, which correctly classified 28 of the 31 prediction samples.
The results obtained in this work indicate that the proposed method is a promising alternative for classifying cereal bar samples. Moreover, the analytical measurement can be performed quickly, without re-agents and involves less consumption of the sample.
Future works could investigate the combination of NIR with other methods, such as soft independent modelling of class analogy (SIMCA), partial least square—discriminant analysis (PLS-DA), arti-ficial neural network (ANN) and k-nearest neighbor (KNN) for the purpose of improving the classification outcome.
Table 2
Classification results obtained with LDA/SW, LDA/SPA and LDA/GA models in the pre-diction set. (1): diet; (2): conventional and (3): Light. The numbers of wavenumbers employed in the LDA/SW, LDA/SPA and LDA/GA models were 4, 12 and 17, respectively. The numbers of prediction samples employed in this study were 9, 11 and 11 for diet, conventional and light types, respectively.
Prediction Set
LDA/SWa LDA/SPA LDA/GA Predicted
type index type Predicted index type Predicted index True type index 1 2 3 CCR 1 2 3 CCR 1 2 3 CCR 1 5 2 2 7 1 1 7 1 1 2 1 7 3 61% 1 8 2 81% - 10 1 90% 3 1 3 7 1 - 10 - - 11 a
The threshold value selected was 0.90.
Fig. 4.DF2 × DF1 score plots for the overall data set (121 samples) using wavenumbers selected by (a) SW, (b) SPA and (c) GA. ( :conventional; : diet and :light).
Acknowledgments
This work was supported by NUQAAPE (FACEPE) and CNPq.
References
Adolfo Lutz Institute (1985).Métodos químicos e físicos para análise de alimentos (3 ª ed.). (São Paulo).
Association of Official Analytical Chemists—AOAC ().Official methods of analysis of the association of official analytical chemists (16th ed.). Washington, D.C.: AOAC International.
Barnes, R. J., Dhanoa, M. S., & Lister, S. J. (1989). Standard normal variate transforma-tion and de-trending of near-infrared diffuse reflectance spectra.Applied Spectros-copy,43, 772–777.
Berrueta, L. A., Alonso-Salces, R. M., & Héberger, K. (2007). Supervised pattern recogni-tion in food analysis.Journal of Chromatography. A,1158, 196–214.
Bingol, G., Zhang, A., Pan, Z., & McHugh, T. H. (2012). Producing lower-calorie deep fat fried French fries using infrared dry-blanching as pretreatment.Food Chemistry,
132, 686–692.
Brazilian Agency of Sanitary Surveillance (2012). Portaria ANVISA SVS/MS No. 27 e 29 de 13.01.1998. Available athttp://www.anvisa.gov.br/alimentos/legis/especifica/ regutec.htm(Accessed in 24.08.12).
Brereton, R. G. (2009).Chemometrics for pattern recognition.Bristol, UK: John Wiley & Sons Ltd.
Caneca, A. R., Pimentel, M. F., Galvão, R. K. H., Matta, C. E., Carvalho, F. R., Raimundo, I. M., Jr., et al. (2006). Assessment of infrared spectroscopy and multivariate techniques for monitoring the service condition of diesel-engine lubricating oils.Talanta, 70, 344–352.
Carrillo, E., Varela, P., & Fiszman, S. (2012). Packaging information as a modulator of consumers' perception of enriched and reduced-calorie biscuits in tasting and non-tasting tests.Food Quality and Preference,25, 105–115.
Cen, H., & He, Y. (2007). Theory and application of near infrared reflectance spectroscopy in determination of food quality.Trends in Food Science and Technology,18, 72–83. Cruz, A. G., Cadena, R. S., Alvaro, M. B. V. B., Sant'Ana, A. S., Oliveira, C. A. F., Faria, J. A. F.,
et al. (2013). Assessing the use of different chemometric techniques to discrimi-nate low-fat and full-fat yogurts.LWT- Food Science and Technology,50, 210–214. Egert, S., Wolffram, S., Schulze, B., Langgunth, P., Hubbermann, E. M., Schwarz, K., et al.
(2012). Enriched cereal bars are more effective in increasing plasma quercetin compared with quercetin from powder-filled hard capsules.The British Journal of Nutrition,107, 539–546.
Farinazzi-Machado, F. M. V., Barbalho, S. M., Oshiiwa, M., Goulart, R., & Pessan Junior, O. (2012). Use of cereal bars with quinoa (Chenopodium quinoaW.) to reduce risk fac-tors related to cardiovascular diseases.Ciência e Tecnologia de Alimentos,32, 239–244. Filho, P. A. C. (2009). Rapid determination of sucrose in chocolate mass using near
in-frared spectroscopy.Analytica Chimica Acta,631, 206–211.
Fisher, R. A. (1936). The use of multiple measurements in taxonomic problems.Annals of Eugenics,7, 179–188.
Fonseca, R. S., Santo, V. R. D., Souza, G. B., & Pereira, C. A. M. (2011). Elaboração de barra de cereais com casca de abacaxi.Archivos Latino Americanos Nutrición,61, 216–223. Gambarra Neto, F. F., Marino, G., Araujo, M. C. U., Galvão, R. K. H., Pontes, M. J. C., Medeiros, E. P., et al. (2009). Classification of edible vegetable oils using square wave voltammetry with multivariate data analysis.Talanta,77, 1660–1666. Gerend, M. A. (2009). Does calorie information promote lower calorie fast food choices
among college students?The Journal of Adolescent Health,44, 84–86.
Gori, A., Maggio, R., Cerretani, L., Nocetti, M., & Caboni, M. F. (2012). Discrimination of grated cheeses by Fourier transform infrared sprectroscopy coupled with chemometric techniques.International Dairy Journal,23, 115–120.
Granato, D., Branco, G. F., Faria, J. A. F., & Cruz, A. G. (2011). Characterization of Brazilian lager and brown ale beers based on color, phenolic compounds, and antioxidant activity using chemometrics.Journal of the Science of Food and Agriculture,91, 563–571.
Heenan, S., Soukoulis, C., Silcock, P., Fabris, A., Aprea, E., Cappellin, L., et al. (2012). PTR-TOF-MS monitoring of in vitro and in vivoflavour release in cereal bars with varying sugar composition.Food Chemistry,131, 477–484.
Kennard, R. W., & Stone, L. A. (1969). Computer aided design of experiments.
Technometrics,11, 137–148.
Kocken, P., Buijs, G., & Snel, J. (2006). Promotion of purchase of low-calorie foods from vending machines at secondary schools in the Netherlands.European Journal of Public Health,16, 174-174.
Lobato, L. P., Pereira, A. E. C., Lazaretti, M. M. L., Barbosa, D. S., Carreira, C. M., Mandarino, J. M. G., et al. (2012). Snack bars with high soy protein and isoflavone content for use in diets to control dyslipidaemia.International Journal of Food Sciences and Nutrition,63, 49–58.
Moreira, E. D., Pontes, M. J. C., Galvão, R. K. H., & Araújo, M. C. U. (2009). Near infrared reflectance spectrometry classification of cigarettes using the successive projec-tions algorithm for variable selection.Talanta,79, 1260–1264.
Pasquini, C. (2003). Near infrared spectroscopy: Fundamentals, practical aspects and analytical applications.Journal of the Brazilian Chemical Society,14, 198–219. Pontes, M. J. C., Galvão, R. K. H., Araújo, M. C. U., Moreira, P. N. T., Pessoa Neto, O. D., José,
G. E., et al. (2005). The successive projections algorithm for spectral variable selection in classification problems.Chemometrics and Intelligent Laboratory Systems,78, 11–18. Riovanto, R., Cynkar, W. U., Berzaghi, P., & Cozzolino, D. (2011). Discrimination be-tween Shiraz wines from different Australian regions: The role of spectroscopy and chemometrics.Journal of Agricultural and Food Chemistry,59, 10356–10360. Roggo, Y., Duponchel, L., & Huvenne, J. -P. (2003). Comparison of supervised pattern
recognition methods with McNemar's statistical test application to qualitative analysis of sugar beet by near-infrared spectroscopy.Analytica Chimica Acta,477, 187–200.
Rohman, A., & Che Man, Y. B. (2011). The use of Fourier transform mid infrared (FT-MIR) spectroscopy for detection and quantification of adulteration in virgin coconut oil.Food Chemistry,129, 583–588.
Santos, C. T., Bonomo, R. F., Fontan, R. C. I., Bonomo, P., Veloso, C. M., & Fontan, G. C. R. (2011). Characterization and sensorial evaluation of cereal bars with jackfruit.Acta Scientiarum—Technology,33, 81–85.
Savitzky, A., & Golay, M. J. E. (1964). Smoothing and differentiation of data by
simpli-fied least squares procedures.Analytical Chemistry,36, 1627–1639.
Scott, R., Donoghoe, M., Watts, G. F., O'Brien, R., Pardy, C., Taskinen, M. R., et al. (2011). Impact of metabolic syndrome and its components on cardiovascular disease event rates in 4900 patients with type 2 diabetes assigned to placebo in thefield randomised trial.Cardiovascular Diabetology,10, 102.
Silva, A. C., Pontes, L. F. B. L., Pimentel, M. F., & Pontes, M. J. C. (2012). Detection of adul-teration in hydrated ethyl alcohol fuel using infrared spectroscopy and supervised pattern recognition methods.Talanta,93, 129–134.
Silva, C. S., Borba, F. S. L., Pimentel, M. F., Pontes, M. J. C., Honorato, R. S., & Pasquini, C. (2012). Classification of blue pen ink using infrared spectroscopy and linear discrim-inant analysis.Microchemical Journal.http://dx.doi.org/10.1016/j.microc.2012.03.025. Sinelli, N., Cerretani, L., Di Egidio, V., Bendini, A., & Casiraghi, E. (2010). Application of near (NIR) infrared and mid (MIR) infrared spectroscopy as a rapid tool to classify extra virgin olive oil on the basis of fruity attribute intensity.Food Research Interna-tional,43, 369–375.
Souto, U. T. C. P., Pontes, M. J. C., Silva, E. C., Galvão, R. K. H., Araújo, M. C. U., Sanches, F. A. C., et al. (2010). UV–Vis spectrometric classification of coffees by SPA–LDA.Food Chemistry,119, 368–371.
Souza, S. S., Cruz, A. G., Walter, E. H. M., Faria, J. A. F., Celeghini, R. M. S., Ferreira, M. M. C., et al. (2011). Monitoring the authenticity of Brazilian UHT milk: A chemometric approach.Food Chemistry,124, 692–695.
Vaid, T. P., Burl, M. C., & Lewis, N. S. (2001). Comparison of the performance of different discriminant algorithms in analyte discrimination tasks using an array of carbon black-polymer composite vapor detectors.Analytical Chemistry,73, 321–331. Villavicencio, A. L. C. H., Araújo, M. M., Fanaro, G. B., Rela, P. R., & Mancini-Filho, J.
(2007). Sensorial analysis evaluation in cereal bars preserved by ionizing radiation processing.Radiation Physics and Chemistry,76, 1875–1877.
Workman, J. J. (1996). Interpretive spectroscopy for near infrared.Applied Spectroscopy Reviews,31, 251–320.
Workman, J., & Weyer, L. (2008).Practical guide to interpretive near—infrared spectros-copy.New York: CRC Press.
Zaveri, S., & Drummond, S. (2009). The effect of including a conventional snack (cereal bar) and a nonconventional snack (almonds) on hunger, eating frequency, dietary intake and body weight.Journal of Human Nutrition and Dietetics,22, 461–468.
