L'âge a de tout temps constitué un attribut identitaire important. Nous avons développé au fil de l'évolution une aptitude innée à classer les individus en fonction de leur âge. Cette classification s'appuie en grande partie sur le visage et sur les transformations anatomiques qu'il subit au cours du temps. De plus en plus de traitements cosmétiques, dermatologiques et d'interventions chirurgicales s'attaquant à un signe ou un groupe de signes spécifiques du vieillissement sont mis en oeuvre pour annuler, ou tout au moins masquer partiellement l'effet du temps sur le visage. On peut dès lors s'interroger sur l'influence de chacun des signes sur notre capacité à prédire l'âge d'un individu en observant son visage. Afin de construire un algorithme capable de déterminer l'âge d'individus à partir de leurs photos, nous nous sommes intéressés aux signes du vieillissement et à leur impact sur l'âge apparent. Dans un premier temps, nous avons déterminé et analysé les transformations anatomiques qui altèrent le visage à partir de l'âge adulte (au-delà de 20 ans). Puis nous avons étudié les signes sur lequel on se base pour prédire l'âge d'une personne. Enfin, nous avons construit et validé un modèle prédictif de l'âge en s'appuyant sur les observations précédentes. Transformations anatomiques du visage avec l'âge : La prévalence d'un certain nombre de signes de vieillissement (rides, tâches brunes, forme du visage...) a été mesurée sur un panel représentatif de femmes volontaires âgées de 20 à 74 ans. Ces données ont permis d'établir la cinétique d'apparition de ces signes. Appréciation subjective de l'âge: Il s'agissait de déterminer les signes sur lesquels un observateur s'appuie lorsqu'il évalue l'âge d'un sujet. Pour ce faire, nous avons demandé à un panel constitué de 48 observateurs d'attribuer un âge aux volontaires sur lesquelles nous avions précédemment mesuré les signes du vieillissement. Nous avons confirmé avec ce groupe d'observateurs que la perception de l'âge est liée au sexe et à l'âge de l'observateur. De plus, à l'aide d'une régression PLS (Partial Least Square régression), nous avons établi des relations entre les signes du vieillissement et l'âge observé et démontré que selon que l'on soit jeune ou âgé, un homme ou une femme, on n'exploite pas les mêmes signes de vieillissement pour prédire l'âge.Modèle de prédiction : Enfin, nous avons proposé un modèle s'appuyant sur la régression PLS pour prédire automatiquement l'âge à partir des photos du visage. Ce modèle présente la particularité d'associer, dans une approche unifiée, les signes relatifs à la couleur, à la forme et à la texture du visage, à l'âge des sujets. A l'instar des Modèles Actifs D'apparence (AAM), le modèle construit vise à réduire fortement l'information portée par l'ensemble des pixels du visage. Toutefois, ce dernier est supervisé : Il est donc très approprié dans notre contexte puisque que l'on peut mettre en oeuvre une procédure d'apprentissage pilotée par le but. Les performances sont de fait comparables à celles des humains.

This study starts with the review of the literature in customer satisfaction, customer satisfaction methods and models. After selecting a proper customer satisfaction method and model, the study conducts a survey and a questionnaire among the customers and professionals in the modular kitchen sector. The aim of the study is to analyze the factors affecting customer satisfaction and finding out the ones related with the modular kitchen sector. After applying the survey, the relations between the inputs and outputs of the satisfaction are analyzed with the overall satisfaction itself. The strong and weak factors are determined and a proper CRM tool is build-up to realize a decision-support and forecast tool in the study, which can be seen as a beginning for the companies in the real sector in this business to build a much more detailed and ERP integrated software and to use them. The results of the survey are compared with the similar studies from the literature.

Neste trabalho, foram estudadas estratégias para caracterização de bebidas à base de soja (BBS), por meio de análises por espectroscopia no infravermelho médio com transformada de Fourier com acessório de reflexão total atenuada (FTIR-ATR). Foram utilizadas 20 amostras comerciais de BBS, de 7 diferentes sabores e 3 diferentes marcas. Os teores estudados nas BBS foram glicídios totais, glicídios redutores, glicídios não redutores, e proteínas totais. Os modelos de regressão multivariada foram construídos por mínimos quadrados parciais (PLS), empregando como seleção de variáveis os métodos de mínimos quadrados parciais por intervalo (iPLS) e mínimos quadrados parciais por sinergismo de intervalos (siPLS). As seleções de variáveis por siPLS apresentaram os melhores resultados para os modelos construídos. Entre as propriedades avaliadas, a de glicídios totais apresentou modelos com erros de calibração e previsão (RMSECV e RMSEP) baixos, e coeficientes de determinação (R2cv e R2prev) próximos de um. Para proteínas totais, os modelos apresentaram resultados promissores, pois também tiveram erros de calibração e previsão (RMSECV e RMSEP) baixos, e coeficientes de determinação (R2cv e R2prev) próximos de um, considerando-se que as amostras reais e não apresentavam uma variabilidade de concentração de proteínas ideal. Para as propriedades de glicídios redutores e glicídios não redutores, não foram obtidos bons resultados para os modelos de regressão. Desta forma, a metodologia proposta apresenta potencial em análises de rotinas para determinação simultânea de glicídios totais e proteínas, atendendo aos requisitos referente às informações nutricionais na rotulagem das BBS, somando-se às vantagens da espectroscopia no infravermelho, tais como rapidez na análise, elevada frequência analítica, pequena quantidade de amostra necessária, baixo custo, não ser destrutiva e ser ambientalmente amigável. / In this work, strategies were studied for the characterization of soy-based beverages (SBB), by means of Fourier transform infrared spectroscopy with attenuated total reflectance (FTIR-ATR). Twenty commercial samples of SBB were used, of 7 different flavors and 7 different brands. The contents studied in SBB were total sugar, reducing sugar, non-reducing sugars, and total proteins. The multivariate regression models were constructed by partial least squares (PLS), with evaluation of the methods by interval partial least squares (iPLS) and by sinergy interval partial least squares (siPLS), for selection of variables. The selections of variables per siPLS presented the best results for the constructed models. Among the evaluated properties, the total sugar content presented models with low calibration and prediction errors (RMSECV and RMSEP), and determination coefficients (R2cv and R2prev) close to one. For total proteins, the models presented promising results, as they also had low calibration and prediction errors (RMSECV and RMSEP), and determination coefficients (R2cv and R2prev) close to one, considering that the actual samples did not present an ideal protein concentration variability. For the properties of reducing sugars and non-reducing sugars, good results were not obtained for the regression models. In this way, the proposed methodology presents potential in routine analysis for simultaneous determination of total glycogen and protein, taking into account the requirements referring to the nutritional information in the SBB labeling, adding to the advantages of the infrared spectroscopy, such as speed in the analysis, high analytical frequency, small amount of sample required, low cost, non destructive and environmentally friendly.

Neste trabalho, foram estudadas estratégias para caracterização de bebidas à base de soja (BBS), por meio de análises por espectroscopia no infravermelho médio com transformada de Fourier com acessório de reflexão total atenuada (FTIR-ATR). Foram utilizadas 20 amostras comerciais de BBS, de 7 diferentes sabores e 3 diferentes marcas. Os teores estudados nas BBS foram glicídios totais, glicídios redutores, glicídios não redutores, e proteínas totais. Os modelos de regressão multivariada foram construídos por mínimos quadrados parciais (PLS), empregando como seleção de variáveis os métodos de mínimos quadrados parciais por intervalo (iPLS) e mínimos quadrados parciais por sinergismo de intervalos (siPLS). As seleções de variáveis por siPLS apresentaram os melhores resultados para os modelos construídos. Entre as propriedades avaliadas, a de glicídios totais apresentou modelos com erros de calibração e previsão (RMSECV e RMSEP) baixos, e coeficientes de determinação (R2cv e R2prev) próximos de um. Para proteínas totais, os modelos apresentaram resultados promissores, pois também tiveram erros de calibração e previsão (RMSECV e RMSEP) baixos, e coeficientes de determinação (R2cv e R2prev) próximos de um, considerando-se que as amostras reais e não apresentavam uma variabilidade de concentração de proteínas ideal. Para as propriedades de glicídios redutores e glicídios não redutores, não foram obtidos bons resultados para os modelos de regressão. Desta forma, a metodologia proposta apresenta potencial em análises de rotinas para determinação simultânea de glicídios totais e proteínas, atendendo aos requisitos referente às informações nutricionais na rotulagem das BBS, somando-se às vantagens da espectroscopia no infravermelho, tais como rapidez na análise, elevada frequência analítica, pequena quantidade de amostra necessária, baixo custo, não ser destrutiva e ser ambientalmente amigável. / In this work, strategies were studied for the characterization of soy-based beverages (SBB), by means of Fourier transform infrared spectroscopy with attenuated total reflectance (FTIR-ATR). Twenty commercial samples of SBB were used, of 7 different flavors and 7 different brands. The contents studied in SBB were total sugar, reducing sugar, non-reducing sugars, and total proteins. The multivariate regression models were constructed by partial least squares (PLS), with evaluation of the methods by interval partial least squares (iPLS) and by sinergy interval partial least squares (siPLS), for selection of variables. The selections of variables per siPLS presented the best results for the constructed models. Among the evaluated properties, the total sugar content presented models with low calibration and prediction errors (RMSECV and RMSEP), and determination coefficients (R2cv and R2prev) close to one. For total proteins, the models presented promising results, as they also had low calibration and prediction errors (RMSECV and RMSEP), and determination coefficients (R2cv and R2prev) close to one, considering that the actual samples did not present an ideal protein concentration variability. For the properties of reducing sugars and non-reducing sugars, good results were not obtained for the regression models. In this way, the proposed methodology presents potential in routine analysis for simultaneous determination of total glycogen and protein, taking into account the requirements referring to the nutritional information in the SBB labeling, adding to the advantages of the infrared spectroscopy, such as speed in the analysis, high analytical frequency, small amount of sample required, low cost, non destructive and environmentally friendly.

Despite the recent decreasing trend of most traditional types of crime, online property crime (OPC), referring to crime committed online with a financial orientation such as online frauds, scams, and phishing, continues to increase. According to the Internet Crime Complaint Center, the number of reported complaints about OPC have increased by approximately sixteen fold from 16,838 cases in 2000 to 288,012 cases in 2015, and referred financial losses have also increased about sixty times from $17.8 million in 2001 to $1 billion in 2015. The increase in OPC might be directly related to advanced online accessibility due to the accelerated progress of information and communication technology (ICT). Since the progress of ICT continues forward and the advanced ICT infrastructure can affect our routine activities more significantly, issues regarding OPC may become more various and prevalent. The present study aims to explore a macro-social criminogenic structure of OPC perpetration. Specifically, this study focused on exploring probable macro-social predictors of OPC rates and examining how effectively these possible macro-social predictors account for variance in OPC perpetration rates. In addition, this study explored possible predictors of macro-level online opportunity structure, which is expected to have a direct relationship with OPC rates. It also examined how much variance in online opportunity structure was explained by the included possible predictors. With these research purposes, the current study analyzed state-level data of the fifty states in the U.S. by applying a partial least square regression (PLSR) approach. The results indicated that predictors related to macro-social economic conditions such as economic inequality, poverty, economic social support, and unemployment had a significant association with OPC. As expected, indicators in the domain of economic inequality predicted greater OPC rates and those in the domain of economic social support were related to lower OPC rates. However, poverty and unemployment predictors were negatively associated with OPC, which is the opposite direction of the relationships between these predictors and traditional street crime. In addition, indicators of online opportunity structure were found to have a significantly positive relationship to OPC as expected. The PLSR model for predicting OPC applied in the current study accounted for approximately 50% of variance in OPC rates across states. For predictors of online opportunity structure, the results indicated that online opportunity was associated with state-level economic and socio-demographic characteristics. States with less poverty, more urban population, and more working age adults were more likely to report more online opportunities. The PLSR model for predicting online opportunity structure explained about 80% of variance in measured online opportunity. These results may imply that some types of macro-social conditions may have an indirect effect on OPC through online opportunity structure as well as their direct effects on OPC. Future study should pay more attention to examining structural relationships of macro-social contexts, online opportunity structure, and OPC to understand macro-level criminogenic mechanism of OPC.

Notre objectif est d’étudier les modalités d’accouchement au sein de la lignée humaine. Pour cela, nous utilisons deux approches complémentaires : la simulation numérique de l’accouchement et l’analyse discriminante des modalités d’accouchement au sein d’un échantillon obstétrical. Dans un premier temps, nous construisons des maillages de bassins et de crânes de foetus fossiles grâce à une méthode d’interpolation : le krigeage. Les groupes fossiles considérés sont les Australopithèques, les premiers représentants du genre Homo (PRGH) et les représentants du genre Homo au Pléistocène moyen et supérieur (RPMS). Les dimensions des crânes juvéniles sont utilisées pour estimer « à rebours » les dimensions néonatales à l’aide de courbes de croissance humaine et de chimpanzé. Nous réalisons une simulation numérique de l’accouchement à partir des maillages de ces dyades « virtuelles ». Puis nous réalisons des analyses discriminantes avec un jeu de données issu de mesures réalisées sur le pelviscanner de femmes et sur les mesures du crâne de leur nouveau-né afin de séparer les modalités d’accouchement grâce aux variables foeto-pelviennes. Ces mêmes variables foeto-pelviennes sont mesurées chez les dyades fossiles afin d’identifier, par les analyses discriminantes, leurs modalités d’accouchement les plus probables. Nos résultats suggèrent un accouchement eutocique sans rotation intra-pelvienne chez les Australopithèques, eutocique avec rotation intrapelvienne chez les PRGH, dystocique ou eutocique chez les RPMS, l’accouchement eutocique est caractérisé par une rotation et une incurvation de la trajectoire de descente. / The purpose of this thesis is to estimate delivery outcomes for extinct hominids. We therefore use two complementary methods : numerical simulation of childbirth and discriminant analysis of delivery outcomes from an obstetrical sample. First, we use kriging to construct meshes of pelves and neonatal skulls. Fossil hominid specimens included in the study are Australopithecines, early Homo (EH) and middle to early Pleistocene Homo (MEPH). We estimate fetal cranial dimensions with chimpanzee or human cranial growth curve that we reversly use and apply on juveniles skull measurements. “Virtual” dyads are formed from pelves and neonatal skulls. Then, we simulate childbirth of these « virtual » dyads. Different levels of laxity of the sacro-iliac junction and different positions of the fetal head are considered. Finally, we use an obstetrical sample: delivery outcome is noted, CT-scans are used to obtain maternal pelvic measurements and diameters of the fetal head were also measured after delivery. A discriminant analysis is performed using this obstetrical sample to separate delivery outcomes thanks to fetal-pelvic measurements. Fossil dyads were subsequently added in the discriminant analysis to assess delivery outcomes to which they belong. Results suggest small fetal-pelvic constraint for Austalopithecines. This constraint is moderate for EH. Fetal-pelvic constraint is more important for MEPH. We suggest that rotational birth appears with EH. The curved trajectory of the fetal head appears with MEPH. Emergence of rotational birth and curved trajectory of fetal head are probably explained by two major increases in brain size during late and middle Pleistocene.

Neste trabalho, foram estudadas estratégias para caracterização de bebidas à base de soja (BBS), por meio de análises por espectroscopia no infravermelho médio com transformada de Fourier com acessório de reflexão total atenuada (FTIR-ATR). Foram utilizadas 20 amostras comerciais de BBS, de 7 diferentes sabores e 3 diferentes marcas. Os teores estudados nas BBS foram glicídios totais, glicídios redutores, glicídios não redutores, e proteínas totais. Os modelos de regressão multivariada foram construídos por mínimos quadrados parciais (PLS), empregando como seleção de variáveis os métodos de mínimos quadrados parciais por intervalo (iPLS) e mínimos quadrados parciais por sinergismo de intervalos (siPLS). As seleções de variáveis por siPLS apresentaram os melhores resultados para os modelos construídos. Entre as propriedades avaliadas, a de glicídios totais apresentou modelos com erros de calibração e previsão (RMSECV e RMSEP) baixos, e coeficientes de determinação (R2cv e R2prev) próximos de um. Para proteínas totais, os modelos apresentaram resultados promissores, pois também tiveram erros de calibração e previsão (RMSECV e RMSEP) baixos, e coeficientes de determinação (R2cv e R2prev) próximos de um, considerando-se que as amostras reais e não apresentavam uma variabilidade de concentração de proteínas ideal. Para as propriedades de glicídios redutores e glicídios não redutores, não foram obtidos bons resultados para os modelos de regressão. Desta forma, a metodologia proposta apresenta potencial em análises de rotinas para determinação simultânea de glicídios totais e proteínas, atendendo aos requisitos referente às informações nutricionais na rotulagem das BBS, somando-se às vantagens da espectroscopia no infravermelho, tais como rapidez na análise, elevada frequência analítica, pequena quantidade de amostra necessária, baixo custo, não ser destrutiva e ser ambientalmente amigável. / In this work, strategies were studied for the characterization of soy-based beverages (SBB), by means of Fourier transform infrared spectroscopy with attenuated total reflectance (FTIR-ATR). Twenty commercial samples of SBB were used, of 7 different flavors and 7 different brands. The contents studied in SBB were total sugar, reducing sugar, non-reducing sugars, and total proteins. The multivariate regression models were constructed by partial least squares (PLS), with evaluation of the methods by interval partial least squares (iPLS) and by sinergy interval partial least squares (siPLS), for selection of variables. The selections of variables per siPLS presented the best results for the constructed models. Among the evaluated properties, the total sugar content presented models with low calibration and prediction errors (RMSECV and RMSEP), and determination coefficients (R2cv and R2prev) close to one. For total proteins, the models presented promising results, as they also had low calibration and prediction errors (RMSECV and RMSEP), and determination coefficients (R2cv and R2prev) close to one, considering that the actual samples did not present an ideal protein concentration variability. For the properties of reducing sugars and non-reducing sugars, good results were not obtained for the regression models. In this way, the proposed methodology presents potential in routine analysis for simultaneous determination of total glycogen and protein, taking into account the requirements referring to the nutritional information in the SBB labeling, adding to the advantages of the infrared spectroscopy, such as speed in the analysis, high analytical frequency, small amount of sample required, low cost, non destructive and environmentally friendly.

Near infrared spectroscopy (NIR) has become increasingly widespread throughout various fields as an alternative method for efficiently phenotyping crops and plants at rates unparalleled by conventional means. With growing reliability, the convergence of NIR spectroscopy and modern machine learning represent a promising methodology offering unprecedented access to rapid, high throughput phenotyping at negligible costs, representing prospects that excite agronomists and plant physiologists alike. However, as is true of all emergent methodologies, progressive refinement towards optimization exposes potential flaws and raises questions, one of which is the cornerstone of this study. Spectroscopic determination of plant functional traits utilizes plants' morphological and biochemical properties to make predictions, and has been validated at the community (inter-family) and individual crop (intraspecific) levels alike, yielding equally reliable predictions at both scales, yet what lies amid these poles on the spectrum of taxonomic scale remains unexplored territory. In this study, we replicated the protocol used in studies of the aforementioned taxonomic scale extremes and applied it to an intermediate scale. Interestingly, we found that predictive models built upon hyperspectral reflectance data collected across three genera of woody plants: Cornus, Rhododendron, and Salix, yielded inconsistent predictions of varying accuracy within and across taxa. Identifying the potential cause(s) underlying variability in predictive power at this intermediate taxonomic scale may reveal novel properties of the methodology, potentially permitting further optimization through careful consideration.

Soil fertility is an important component of forest ecosystem, yet evaluating soil fertility remains one of the least understood aspects of forest science. Phytocentric and geocenctric approaches were used to assess soil fertility in loblolly pine plantations throughout their geographic range in the United States. The model to assess soil fertility using a phytocentric approach was constructed using the relationship between site index and aboveground productivity. Geocentric models used physical and chemical properties of the A-horizon. Soil geocentric models were constructed using two modeling approaches. In the first approach, ordinary least squares methods of multiple regression were used to derive soil fertility estimated from site index using soil physical and chemical properties from the A-horizon. Ordinary least squares methods were found unsuitable due to multicollinearity among the soil variables. In the second approach, a multivariate modeling approach, partial least squares regression, was used to mitigate multicollinearity effects. The best model to quantify soil fertility using soil physical and chemical properties included N, Ca, Mg, C, and sand percentage as the significant predictors. The 3-PG process-based model was evaluated for simulating the response of loblolly pine to changes in soil fertility. Fertility rating (FR) is a parameter in 3-PG that scales soil fertility in the range of 0 to 1. FR values estimated from phytocentric and geocentric approaches were tested against observed production. The 3-PG model prediction of aboveground productivity described 89% percent of the variation in observed aboveground productivity using FR derived from site index and 84% percent of the vari- ation in observed aboveground productivity using FR derived from physical and chemical properties of the A-horizon. A response function to model dynamics of FR (∆FR) due to one time midrotatoin fertilization of N and P was developed using the Weibull function. The magnitude of ∆FR varied with intensity of N and time since application of fertilizer. The hypothesis that repeated fertilization with N and P eliminate major nutrient deficiency in the southeastern US was tested and a relationship between baseline fertility rating and fertilizer response was developed. An inverse relationship was observed between fertilizer response and baseline FR. / Ph. D.

Fourier Transform Infrared (FTIR) microscopy is a sensitive method for detecting difference in the morphology of biological cells. In this study FTIR spectra were obtained for uninfected cells, and cells infected with two different viruses. The spectra obtained are difficult to discriminate visually. Here we apply advanced statistical methods to the analysis of the spectra, to test if such spectra are useful for diagnosing viral infections in cells. Logistic Regression (LR) and Partial Least Squares Regression (PLSR) were used to build models which allow us to diagnose if spectral differences are related to infection state of the cells. A three-fold, balanced cross-validation method was applied to estimate the shrinkages of the area under the receiving operator characteristic curve (AUC), and specificities at sensitivities of 95%, 90% and 80%. AUC, sensitivity and specificity were used to gauge the goodness of the discrimination methods. Our statistical results shows that the spectra associated with different cellular states are very effectively discriminated. We also find that the overall performance of PLSR is better than that of LR, especially for new data validation. Our analysis supports the idea that FTIR microscopy is a useful tool for detection of viral infections in biological cells.