Dans cette étude, nous développons et testons un modèle conceptuel de généralisation de la confiance dans les réseaux de franchise à travers les concepts de contrat relationnel et de socialisation organisationnelle. Nous montrons que la confiance peut être généralisée dans un réseau de franchise à travers des mécanismes institutionnels. D’une part, le caractère relationnel du contrat de franchise, gouverné par des normes d’intégrité de rôle, de solidarité, de flexibilité et d’équité, favorise l’émergence de la confiance entre le franchiseur et chaque franchisé. D’autre part, un processus de socialisation institutionnalisée, se traduisant par le partage de valeurs essentielles, le développement d’un sentiment d’appartenance à une même entité et une culture organisationnelle, constitue un vecteur de sentiment de confiance partagé par les membres de réseaux de franchise. Les conséquences de la généralisation de la confiance sont principalement un engagement plus fort des membres, une plus grande propension à coopérer et une plus forte performance individuelle et collective. / In this study, we develop and test a model of trust generalization in franchise networks based on the relational contract theory and organizational socialization concept. We show that trust can emerge between franchisor and any franchisee if the franchise contract is seen as relational governed by norms of role integrity, solidarity, flexibility and equity. Trust between franchisees can follow to a process of institutionalized socialization, who conducts to share essentials values, the same vision on the organization, and the emergence of a corporate culture. Altogether, trust generalization in franchise networks comes from institutional mechanisms of relational contract and a process of institutionalized socialization. We assume that the most important consequences of trust generalization in franchise organizations are more commitment of the franchisees, a better cooperation between all members, and a great performance for all.

Partial Least Squares (PLS) is an important statistical technique with multipleand diverse applications, used as an effective regression method for correlated orcollinear datasets or for datasets that are not full rank for other reasons. A shorthistory of PLS is followed by a review of the publications where the issues with theapplication PLS that have been discussed. The theoretical basis of PLS is developedfrom the single value decomposition of the covariance, so that the strong links between principal components analysis and within the various PLS algorithms appear as a natural consequence. Latent variable selection by crossvalidation, permutation and information criteriaare examined. A method for plotting crossvalidation results is proposed that makeslatent variable selection less ambiguous than conventional plots. Novel and practicalmethods are proposed to extend published methods for latent variable selection byboth permutation and information criteria from univariate PLS1 models to PLS2 multivariate cases. The numerical method proposed for information criteria is also more general than the algebraic methods for PLS1 that have been recently published as it does not assume any particular form for the PLS regression coefficients. All of these methods have been critically assessed using a number of datasets, selected specifically to represent a diverse set of dimensions and covariance structures. Methods for simulating multivariate datasets were developed that allow controlof correlation and collinearity in both regressors and responses independently. Thisdevelopment also allows control over the variate distributions. Statistical design ofexperiments was used to generate plans for the simulation that allowed the factorsthat infuence PLS model fit and latent variable selection. It was found that all thelatent variable selection methods in the simulation tend to overfit and the feature inthe simulation that causes overfitting has been identified.

Multivariate statistical process control (MSPC) techniques play an important role in industrial batch process monitoring and control. This research illustrates the capabilities and limitations of existing MSPC technologies, with a particular focus on partial least squares (PLS).In modern industry, batch processes often operate over relatively large spaces, with many chemical and physical systems displaying nonlinear performance. However, the linear PLS model cannot predict nonlinear systems, and hence non-linear extensions to PLS may be required. The nonlinear PLS model can be divided into Type I and Type II nonlinear PLS models. In the Type I Nonlinear PLS method, the observed variables are appended with nonlinear transformations. In contrast to the Type I nonlinear PLS method, the Type II nonlinear PLS method assumes a nonlinear relationship within the latent variable structure of the model. Type I and Type II nonlinear multi-way PLS (MPLS) models were applied to predict the endpoint value of the product in a benchmark simulation of a penicillin batch fermentation process. By analysing and comparing linear MPLS, and Type I and Type II nonlinear MPLS models, the advantages and limitations of these methods were identified and summarized. Due to the limitations of Type I and II nonlinear PLS models, in this study, Neural Network PLS (NNPLS) was proposed and applied to predict the final product quality in the batch process. The application of the NNPLS method is presented with comparison to the linear PLS method, and to the Type I and Type II nonlinear PLS methods. Multi-way NNPLS was found to produce the most accurate results, having the added advantage that no a-priori information regarding the order of the dynamics was required. The NNPLS model was also able to identify nonlinear system dynamics in the batch process. Finally, NNPLS was applied to build the controller and the NNPLS method was combined with the endpoint control algorithm. The proposed controller was able to be used to keep the endpoint value of penicillin and biomass concentration at a set-point.

Les modèles de régression dans le cadre des données censurées supposent comme pour tous modèles de régression que l'on ait plus d'observations que de descripteurs et des descripteurs pas trop corrélés entre eux. Ces hypothèses ne sont pas souvent vérifiées dans la pratique et les approches classiques deviennent caduques. C'est le cas par exemple en pharmacogénomique lorsque l'on cherche à prédire la probabilité de survie de patients à partir de profiles, ou signatures transcriptomiques, construits à partir de l'expression de milliers de gènes. L'objectif de cette thèse est d'apporter une solution à ce problème en utilisant les principes de la régression PLS. Le modèle PLS-Cox qui est proposé est issu d'une généralisation de la régression PLS à tous modèles de régression. Il fournit une alternative régularisée aux modèles de survie en grande dimension (p>>n). L'utilisation d'une reparamétrisation de type « Kernel » des algorithmes PLS a permis en outre de développer des solutions à la fois très rapides et particulièrement utiles en très grande dimension, mais aussi adaptées pour l'analyse de structures non-linéaires des données. Une autre solution à ce problème, rapide et simple à mettre en œuvre, a été développée à partir des résidus de la déviance. Une alternative à PLS-NIPALS a été proposée pour la gestion des valeurs manquantes en introduisant le concept d'imputation multiple en régression PLS simple et généralisée. Finalement, on s'est intéressé à la recherche de modèles plus parcimonieux en proposant une approche de type « Thresholding PLS » qui opère implicitement par sélection des variables.

Les contaminations par les moisissures représentent un problème majeur au sein de l'industrie agroalimentaire, pharmaceutique, cosmétique, et dans le secteur médical. Actuellement, l'identification des champignons filamenteux est basée sur l'analyse des caractéristiques phénotypiques, nécessitant une expertise et pouvant manquer de précision, ou sur les méthodes moléculaires, coûteuses et fastidieuses. Dans ce contexte, l'objectif de cette étude a consisté à développer un protocole simple et standardisé à l'aide de la spectroscopie infrarouge à transformée de Fourier (IRTF) combinée à une méthode d'analyse chimiométrique, proposant une méthode alternative pour l'identification rapide des moisissures. Au total, 498 souches de champignons filamenteux (45 genres et 140 espèces) ont été analysées à l'aide d'un spectromètre IRTF à haut débit. L'analyse discriminante des moindres carrés partiels (PLS -DA), méthode chimiométrique supervisée, a été appliquée à chaque spectre dans les gammes spectrales 3200-2800 et 1800-800 cm-1. Différents modèles de calibration ont été construits à partir de 288 souches, ceci en cascade de la sous-division jusqu'à l'espèce en se basant sur la taxonomie actuelle. La prédiction des spectres en aveugle, obtenus à partir de 105 souches, au niveau du genre et de l'espèce est respectivement de 99,17 % et 92,3 %. La mise en place d'un score de prédiction et d'un seuil a permis de valider 80,22 % des résultats. L'implémentation d'une fonction de standardisation (SF) a permis d'augmenter le pourcentage de spectres bien prédits, acquis sur un autre instrument, de 72,15 % (sans fonction) à 89,13 %, validant la transférabilité de la méthode. Puisqu'une biomasse mycélienne suffisante peut être obtenue après 48h de culture et que la préparation des échantillons implique l'utilisation d'un protocole simple, la spectroscopie IRTF combinée à la PLS-DA apparaît comme une méthode rapide et peu coûteuse, ce qui la rend particulièrement attractive pour l'identification des champignons filamenteux au niveau industriel. Les résultats obtenus placent la spectroscopie IRTF parmi les méthodes analytiques prometteuses et avant-gardistes, possédant un haut pouvoir discriminant et une forte capacité d'identification, en comparaison avec les techniques conventionnelles. / Mold contaminants represent a major problem in various areas such as food and agriculture, pharmaceutics, cosmetics and health. Currently, molds identification is based either on phenotypic characteristics, requiring an expertise and can lack accuracy, or on molecular methods, which are quite expensive and fastidious. In this context, the objective was to develop a simple and standardized protocol using Fourier transform infrared (FTIR) spectroscopy combined with a chemometric analysis, allowing to implement an alternative method for rapid identification of molds. In total, 498 fungal strains (45 genera and 140 species) were analyzed using a high-throughput FTIR spectrometer. Partial Least Squares Discriminant Analysis (PLS-DA), a supervised chemometrics method, was applied to each spectrum in the spectral ranges 3200-2800 and 1800-800 cm-1 for the identification process. Using 288 strains, different calibration models were constructed in cascade and following the current taxonomy, from the subphylum to the species level. Blind prediction of spectra from 105 strains at the genus and species levels was achieved at 99.17 % and 92.3% respectively. The establishment of a prediction score and a threshold permitted to validate 80.22% of the obtained results. The implementation of a standardization function (SF) permitted to increase the percentage of well predicted spectra from strains analyzed using another instrument from 72.15% (without SF) to 89.13% and permitted to verify the transferability of the method. Since sufficient mycelial biomass can be obtained at 48h culture and sample preparation involved a simple protocol, FTIR spectroscopy combined with PLS-DA is a very rapid and cost effective method, which could be particularly attractive for the identification of moulds at the industrial level. The results obtained places FTIR spectroscopy among the avant-garde promising analytical approaches, with high discriminant power and identification capacity, compared to conventional techniques.

Os estudos de impacto ambiental em sistemas aquáticos são de grande importância, tanto no âmbito social quanto no econômico. Estes têm que contemplar a avaliação dos danos produzidos pelos mais variados tipos de substâncias, criando a necessidade de conhecer a ação destas ao sistema em questão. O resultado da interação de substâncias com o ambiente pode ser explorado por meio de experimentos envolvendo o coeficiente de partição octanol/água (Kow). A escolha do Kow se deve ao fato deste se relacionar a interação do composto escolhido com o meio, no que diz respeito à absorção e transporte. No caso deste trabalho, uma série de fenóis foi estudada em termos do coeficiente de partição. Cálculos teóricos foram realizados como alternativa para mensuração experimental do coeficiente de partição octanol/água. A elaboração de modelos QSAR (Quantitative Structure Activity Relationship) para a previsão de valores de coeficiente de partição octanol/água foi realizado por meio de cálculos envolvendo física quântica e quimiometria. Os dados das variáveis foram obtidos para o grupo de fenóis em questão utilizando os métodos de cálculos semiempíricos AM1, PM3, PM6 e PM7. A comparação dos modelos QSAR obtidos permitiu a conclusão de que, para os descritores selecionados, o conjunto de valores obtidos com o método PM7 proporcionou a criação dos melhores modelos. Utilizando como variáveis selecionadas energia eletrônica, área molecular, volume molecular, energia de repulsão core-core e massa molecular, obteve-se um modelo que acumulava aproximadamente 99,99% de toda a informação em três componentes, com um coeficiente de correlação externo (referente à calibração) em torno de 0,93, o que indica a força preditiva do modelo. A relação da toxicidade com o Índice de Qualidade das Águas também foi investigado por meio da relação dos valores LogKow obtidos e pH. Com o pKa de cada fenol proveniente de dados teóricos (e fazendo uso da relação de pKa e pH proposta por Henderson-Hasselbach), pode-se concluir que não existe influência intrínseca dos valores de LogKow nos valores de pH, verificado com o baixo coeficiente de correlação entre as variáveis, em torno de 0,15. / The environmental impact on aquatic systems are of great importance, both in the social sphere as the economic. These must include the assessment of the damage caused by all sorts of substances, creating the need to know the action of these to the system in question. The result of the interaction of substances with the environment can be explored through experiments involving octanol / water partition coefficient (Kow). The choice of Kow is due to relate the interaction of the compound with the chosen medium, as regards the absorption and transport. In the case of this work, a number of phenols was studied in terms of the partition coefficient. Theoretical calculations have been performed as an alternative to experimental measurement of octanol / water partition. The use of models QSAR (Quantitative Structure Activity Relationship) for predicting values of octanol / water partition was made by means of calculations involving quantum physics and chemometrics. The data for each variable were obtained for the group concerned phenols using the semi-empirical calculation methods AM1, PM3, PM6 and PM7. A comparison of the obtained QSAR models allowed the conclusion that, for the selected descriptors, a set of values obtained with the method provided PM7 the creation of the best models. Using as selected electronic energy, molecular area, molecular volume, core-core repulsion energy and molecular mass variables, we obtained a model that accumulated approximately 99.99% of all information in three components, with an external correlation coefficient (referring to the calibration) around 0,93, which shows the predictive power of the model. The relationship of toxicity with the Water Quality Index was also investigated by means of the relation of LogKow values obtained and pH. With the pKa of the phenol from each theoretical data (and using the relationship of pKa and pH proposed by Henderson-Hasselbach), it can be concluded that there is no influence of intrinsic values LogKow in pH checked with low coefficient correlation between the variables, around 0,15.

O uso do LIBS na agricultura tem crescido nos últimos anos, sendo usando na atualidade para avaliar composição elementar de solos e fertilizantes, detectar a presença de metais pesados em solos e água, monitorar a emissão de carbono para a atmosfera e predizer doenças em plantas ainda em suas primeiras fases de desenvolvimento. Esta técnica é considerada bastante promissora na agricultura devido a sua capacidade de detecção elementar rápida, com pouca ou nula necessidade de preparação da amostra e por não gerar resíduos químicos. Apesar disto, apresenta ainda grandes desafios, principalmente em amostras complexas como o solo, devido a efeitos de matriz, por exemplo. Neste cenário, a determinação das características do sistema LIBS mais adequado para a análise de solo é importante, pois isto permitirá a escolha adequada da configuração a ser usada para se obter resultados de maior acurácia. Assim, nesta tese são comparados dois sistemas experimentais LIBS com lasers de 532 nm (modelo Brilliant, da Quantel, monomodo) e 1064 nm (modelo ULTRA, da Quantel, multimodo), analisando a temperatura e densidade eletrônica do plasma através do método de Saha-Boltzmann, para em seguida verificar a condição de LTE. Para analisar qual sistema teve maior eficiência no processo de ablação se comparou as crateras formadas na amostra, por meio de MEV. A relação entre o sinal e o ruído também foi observada para picos de importantes nutrientes encontrados no solo: carbono (247,84 nm), ferro (244,45 nm), magnésio (280,27 nm) e potássio (766,48 nm). Para verificar o desempenho de ambos os sistemas na determinação quantitativa destes mesmos nutrientes, foram utilizados modelos multivariados através de PLS, como uma alternativa a calibração tradicional com o modelo univariado. Em relação ao sinal/ruído, o sistema de 1064 nm apresentou melhor resultado, a temperatura para ambos os sistemas foi semelhante e o sistema de 532 nm se mostrou mais eficiente no processo de ablação. Contudo, para a quantificação elementar, ambos os sistemas possibilitaram a construção de curvas de calibração lineares com R2 > 0, 85 utilizando PLS, enquanto que utilizando o modelo univariado a correlação foi menor a 0; 8 para todas as curvas de calibração. O RMSEP foi menor para o modelo obtido a partir do sistema de 532 nm para o ferro e carbono, enquanto que para o potássio e magnésio o sistema de 1064 nm apresentou menor RMSEP. Considerando as vantagens econômicas do laser de 1064 nm, este mostrou um excelente custo/benefício, em comparação ao laser de 532 nm, além de ser mais compacto. / The use of LIBS in agriculture has grown in recent years being currently used to evaluate the elementary composition of soils and fertilizers, to detect the presence of heavy metals in soils and water, to monitor the carbon emissions to the atmosphere and to predict diseases in plants, still in its early stages of development. This technique is considering to be very promising in agriculture because of its fast elemental detection capability, with minimum or no sample preparation and no chemical residues generation. Although, it presents great challenges, especially in complex samples such as soil, due to matrix effects, for example. In this scenario, determining the type of the most suitable LIBS system for the analysis for soil is important as this will allows the best choice of the configuration to be used obtaining more accurate results. Thus, in this thesis two experimental LIBS systems are compared with 532 nm (single mode from Quantel, Brilliant) and 1064 nm (multimode from Ultra, Quantel) lasers, analyzing the plasma temperature and electronic density through the Saha-Boltzmann method, and these will verify the lower bound of the LTE condition. In order to analyze which system had the highest efficiency in the ablation process, the craters formed in the sample were compared by MEV. The relationship between the signal and the noise was also observed for peaks of important nutrients found in the soil: carbon (247.84 nm), iron (244,45 nm), magnesium (280,27 nm) and potassium (766,48 nm)). To verify the performance of both systems to the quantitative determination of these same nutrients, the multivariate model PLS was used, as an alternative to traditional calibration with the univariate model. In relation to the signal/noise ratio, the 1064 nm laser had better results, the temperature for both systems was similar and the 532 nm laser shown more efficiency in the ablation process. However, for the elementary quantification, both systems allowed the construction of calibration linear curves with R2 > 0, 85 using PLS, whereas using a univariate model the correlation was lower than 0,8 for any of the calibration curves. The RMSEP was lower for the model obtained with the 532 nm system for iron and carbon, while for potassium and magnesium the 1064 nm laser showed RMSEP lower. Considering the economical advantages of the multimode 1064 nm laser, it shown an excellent cost benefit, in comparision of the single mode 532 nm, besides being more compact.

Com a evolução tecnológica pela qual o mundo vem passando nos últimos anos, o tema Marketing Analytics vem ganhando relevância gerencial e acadêmica. No entanto, pouco se avançou no entendimento do que determina a inclinação de uma empresa a adotar a prática de Analytics para tomada de decisão nas atividades de Marketing, ou seja, sua Orientação para Marketing Analytics (OMA). Além disso, pouco se sabe sobre os impactos dessa orientação nos resultados da empresa. Para suprir essas lacunas, essa pesquisa buscou entender quais os construtos antecedentes que poderiam influenciar a Orientação para Marketing Analytics, bem como se essa orientação levaria a um melhor desempenho do negócio. Para tanto, por meio de uma vasta revisão da literatura, delimitou-se o escopo de Orientação para Marketing Analytics, apresentando uma definição para o construto, um modelo teórico e proposições de pesquisa. Na etapa seguinte, foram feitas entrevistas em profundidade com especialistas em Analytics e executivos de marketing para aprofundar o conceito de Orientação para Marketing Analytics e para entender quais os determinantes para uma empresa ser orientada para Marketing Analytics. A terceira e última parte desse trabalho se focou em desenvolver um modelo de mensuração para Orientação para Marketing Analytics e testar o modelo estrutural proposto junto a uma amostra de 127 profissionais de marketing e especialistas em Analytics. As respostas foram analisadas por meio de Modelagem de Equações Estruturais (PLS-SEM) e os resultados indicaram que o suporte da alta administração desempenha um papel relevante na valorização de habilidades analíticas das pessoas, nos investimentos em infraestrutura tecnológica e na orientação para processos da empresa. Além disso, observou-se que todos esses antecedentes apresentam influência significativa na Orientação para Marketing Analytics da empresa que, por sua vez, tem impacto positivo no desempenho percebido do negócio. / With the technological evolution that the world has been going through in recent years, the topic Marketing Analytics has gained managerial and academic relevance. However, little progress has been made in understanding what determines a company\'s willingness to adopt the practice of Analytics for decision-making in Marketing activities, ie its Marketing Analytics Orientation (MAO). In addition, little is known about the impacts of this orientation on company results. To fill these gaps, this research sought to understand which antecedent could influence the Marketing Analytics Orientation, as well as whether such orientation would lead to better business performance. To do so, through a vast literature review, the scope of Marketing Analytics Orientation was delimited, presenting a definition for the construct, a theoretical model and research propositions. In the next step, in-depth interviews were conducted with analytics experts and marketing executives to deepen the concept of Marketing Analytics Orientation and to understand the determinants for a company to be Marketing Analytics-oriented. The third and final part of this work focused on developing a measurement model for Marketing Analytics Orientation and testing the proposed structural model with a sample of 127 marketing professionals and Analytics experts. The responses were analyzed through Structural Equation Modeling (PLS-SEM) and the results indicated that the top management support plays a relevant role in the appreciation of people\'s analytical skills, on investments in technological infrastructure and in the company\'s processes orientation. The result of this work showed that all of these antecedents had significant influence on the company\'s Marketing Analytics Orientation, which in turn had a positive impact on the perceived performance of the business.

A depressão, também conhecida como distúrbio depressivo maior, é uma doença extremamente séria caracterizada por distúrbios afetivos que apresenta sintomas como tristeza, pessimismo, baixa autoestima, alterações nas funções vegetativas como sono e apetite e cognitivas como concentração, memória e atenção. A depressão afeta hoje cerca de 300 milhões de pessoas sendo essa a doença mais incapacitante do mundo, segundo dados da Organização Mundial da Saúde. O receptor humano 5-HT2a tem sido associado a inúmeras condições neurológicas e moléculas ligantes seletivas a esse receptor podem apresentar potencial terapêutico no tratamento de distúrbios de comportamento, tais como a depressão. Com o propósito de contribuir para o planejamento de novos fármacos mais eficazes no tratamento da depressão realizou-se um estudo de QSAR (do inglês Quantitative structure-activity Relationship) com um conjunto de 106 compostos arilpiperazínicos utilizando os métodos de Mínimos Quadrados Parciais (PLS) e Redes Neurais Artificias (ANN). O modelo de PLS foi obtido com 76 compostos no conjunto de treinamento e 11 compostos no conjunto teste (r2 = 0,749 e q2 = 0,696). Os testes de validação leave-N-out, randomização e detecção de outliers confirmaram a robustez e estabilidade dos modelos e demonstraram que os mesmos não foram gerados através de correlações ao acaso. O modelos de Redes Neurais Artificiais de Perceptron de Multicamadas (MLP-ANN) foi gerado utilizando funções de transferência tansig-tansig. O treinamento das redes demonstrou que o melhor modelo apresentava arquitetura 7-10-1. Tal modelo apresentou valores de erro quadrático médio (EQM) igual a 0,00964, desvio médio absoluto (DMA) igual a 0,0775 e r2treinamento, r2validação e r2teste iguais a 0,794, 0,795 e 0,788, respectivamente. Os descritores dos modelos matemáticos gerados mostraram concordância com as propriedades moleculares dos compostos e os valores de atividade biológica preditos pelos modelos de PLS e ANN demonstraram que a interação dos descritores é satisfatoriamente relacionada tanto por aproximações lineares quanto não-lineares. / Depression, major depressive disorder or clinical depression, is a serious mood disorder characterized by severe symptoms like emotional suffering and feeling miserable that affects vegetative functions as sleeping and eating and cognitive functions like concentration, memory and attention. The world mental health estimates that depression affects more than 300 million people worldwide and is the leading cause of world disability. The human receptor 5-HT2a has been associated to many neurological dysfunctions. Molecules that selectively bind to this receptor present the potential to be used as an effective treatment of mental health disorders, such as depression. Our primary aim in this work is to contribute to the development of new drugs for the treatment of depression. Quantitative Structure-Activity Relationships (QSAR) studies with a series of 106 arylpiperazines that interact with the neuroreceptor 5-HT2a were conducted employing the methods Partial Least Squares (PLS) and Artificial Neural Networks (ANN). The PLS model was obtained with a training set of 76 compounds and a 19 compounds test set (r² = 0. 749, q² = 0.696). Leave-N-out analysis, biological activity randomization and outliers detection confirmed the robustness and stability of the models and showed that they were not obtained by chance correlations. Predictive models were also generated by Multilayer Perceptron Artificial Neural Network (MLP-ANN) trained with tansig-tansig transfer functions showed that the best results were obtained for a 7-10-1 architecture (r²training = 0. 794, r²validation = 0.795 and r²test = 0.788). The descriptors used to construct the predictive models showed good agreement with the arylpiperazines molecular properties, and the biological activity predicted by the PLS and ANN methods suggested that the descriptor interactions can be described by a linear or a nonlinear approach as well.

Neste trabalho estuda-se a aplicação de métodos estatísticos multivariáveis para a classificação de gasolinas comerciais em conformidade à legislação vigente. Atualmente, a ANP baseia a classificação em limites máximos e mínimos para uma série de diferentes propriedades físico-químicas. O objetivo do trabalho é propor uma metodologia para fazer uma triagem das amostras coletadas durante o Programa de Monitoramento da Qualidade dos Combustíveis através de um método de classificação. Ela utiliza a espectroscopia NIR, que é uma técnica rápida e não destrutiva, como método analítico. Com isto será possível reduzir o número de ensaios físico-químicos que não necessariamente seriam realizados sistematicamente em todas as amostras, reduzindo-se os custos e aumentando-se a quantidade de postos monitorados. As análises NIR produzem grandes quantidades de dados, o que leva à utilização de técnicas estatísticas multivariáveis para estabelecer as metodologias de classificação. Neste trabalho utilizam-se técnicas já consagradas, como a PCA e a PLS para a compressão dos dados e a LDA e QDA para a classificação das amostras. Os dados analisados correspondem às propriedades físico-químicas e aos espectros NIR de um conjunto de 216 amostras de gasolinas comerciais, utilizado para a concepção dos modelos de classificação, e de outro de 50 amostras, utilizado para a validação dos modelos. Os modelos testados no trabalho foram as combinações da PCA-LDA, PCA-QDA, PLS-LDA, PLS-QDA, PLS (regressão) e a análise dos gráficos de scores (biplot). Os melhores desempenhos foram obtidos pelos gráficos dos scores, em seguida pela regressão PLS, PLS-QDA, PCA-QDA e PLS-QDA. Existem ainda algumas etapas a serem alcançadas para tornar prática a utilização da classificação de gasolinas comerciais através de NIR, no entanto, a contribuição deste estudo é importante pois permitiu demonstrar a sua viabilidade técnica. / In this work, the application of multivariable statistical methods for the classification of commercial gasoline in accordance to applicable laws in Brazil is studied. In the present, the ANP bases the classification of gasoline on lower and upper bounds defined for a number of physico-chemical properties. The objective of this work is to propose an alternative analysis methodology, that is adequate for making a pre-sorting of the samples collected by the Fuel Quality Monitoring Program through a classification method. This method is based on NIR spectroscopy, that is a fast and non-destructive technique, as the analytical method. In this way, it would be possible to reduce the number of physico-chemical analyses, as it would be possible not to perform them on every sample, reducing costs and increasing the quantity and frequency of gas stations that could be monitored. NIR analyses produce a great quantity of data, that makes the use of multivariable statistical techniques necessary in order to set up classification methodologies. In this work the well-known PCA and PLS techniques are used for data compression, and LDA and QDA analyses for sample classification. The data studied correspond to the physico-chemical properties and NIR spectra of a total of 216 commercial gasoline samples, used for model design, and of a 50 samples, used for validation. The classification methods that are tested are combinations of PCA-LDA, PCA-QDA, PLS-LDA, PLS-QDA, PLS (regression) and data compression scores graphical analysis (biplot). Best performance was obtained with compression scores graphical analysis, followed by PLS regression, PLS-QDA, PCA-QDA and PLS-QDA. There are still some steps to be fulfilled before the usage of commercial gasoline classification through NIR could be practical. However, this study has shown that this methodology is technically feasible.