Classification is a very common task in information processing and important problem in many sectors of science and industry. In the case of data measured as a function of a dependent variable such as time, the most used algorithms may not pattern each of the individual shapes properly, because they are interested only in the choiced measurements. For the reason, the presented paper focuses on the specific techniques that directly address the curve clustering problem and classifying new individuals. The main goal of this work is to develop alternative methodologies through the extension to various statistical approaches, consolidate already established algorithms, expose their modified forms fitted to demands of clustering issue and compare some efficient curve clustering methods thanks to reported extensive simulated data experiments. Last but not least is made, for the sake of executed experiments, comprehensive confrontation of effectual utility. Proposed clustering algorithms are based on two principles. Firstly, it is presumed that the set of trajectories may be probabilistic modelled as sequences of points generated from a finite mixture model consisting of regression components and hence the density-based clustering methods using the Maximum Likehood Estimation are investigated to recognize the most homogenous partitioning. Attention is paid to both the Maximum Likehood Approach, which assumes the cluster memberships to be some of the model parameters, and the probabilistic model with the iterative Expectation-Maximization algorithm, that assumes them to be random variables. To deal with the hidden data problem both Gaussian and less conventional gamma mixtures are comprehended with arranging for use in two dimensions. To cope with data with high variability within each subpopulation it is introduced two-level random effects regression mixture with the ability to let an individual vary from the template for its group. Secondly, it is taken advantage of well known K-Means algorithm applied to the estimated regression coefficients, though. The task of the optimal data fitting is devoted, because K-Means is not invariant to linear transformations. In order to overcome this problem it is suggested integrating clustering issue with the Markov Chain Monte Carlo approaches. What is more, this paper is concerned in functional discriminant analysis including linear and quadratic scores and their modified probabilistic forms by using random mixtures. Alike in K-Means it is shown how to apply Fisher's method of canonical scores to the regression coefficients. Experiments of simulated datasets are made that demonstrate the performance of all mentioned methods and enable to choose those with the most result and time efficiency. Considerable boon is the facture of new advisable application advances. Implementation is processed in Mathematica 4.0. Finally, the possibilities offered by the development of curve clustering algorithms in vast research areas of modern science are examined, like neurology, genome studies, speech and image recognition systems, and future investigation with incorporation with ubiquitous computing is not forbidden. Utility in economy is illustrated with executed application in claims analysis of some life insurance products. The goals of the thesis have been achieved.

A atividade bovina do Brasil é uma das maiores do mundo, devido principalmente, ao uso predominante de pastagens tropicais na dieta do gado, porém, o alto custo dos insumos, com destaque para a adubação nitrogenada, dificulta a obtenção de altas produções de forragens. Por esta razão, a utilização de novas tecnologias que permitam melhorar o manejo e aumentar o rendimento de culturas como o capim \'Mombaça\', por meio do uso racional dos fatores de produção, é de alta relevância. O objetivo deste estudo foi avaliar o uso de dados hiperespectrais para a discriminação de diferentes tratamentos de adubação nitrogenada, e comparar diferentes métodos para obter valores de concentração de clorofila a partir de assinaturas espectrais de capim \'Mombaça\'. Foram estabelecidos quatro tratamentos de com doses diferentes de adubação nitrogenada. Foi medida a reflectância espectral de folhas coletadas em cada tratamento por meio de um sensor hiperespectral em laboratório além de medições de clorofila das folhas. Foi encontrado que as assinaturas espectrais de cada tratamento tiveram comportamentos diferentes, principalmente nas regiões do verde e do red-edge, sendo que tais diferenças dependeram da quantidade de adubo nitrogenado aplicado e do teor de clorofila foliar. A separação dos tratamentos foi possível mediante o uso de análise linear discriminante, baseando-se nos dados de reflectância espectral obtidos em cada tratamento. A obtenção de valores de concentração de clorofila da folha por meio de reflectância espectral foi possível por meio de técnicas de machine learning, destacando-se o support vector machine, como a melhor alternativa. As regiões do red-edge e do verde foram as mais influentes para realizar o cálculo dos valores de concentração de clorofila na folha. / Brazil\'s cattle production is one of the largest in the world, due mainly, to the predominant use of tropical pastures in the cattle diet. However, the high cost of inputs, especially nitrogen fertilizer, makes it difficult to obtain high forage production, for this reason, the use of new technologies to improve the management and to boost the yield of crops such as Mombaça grass, through the rational use of resources, is of high relevance. The aim of this study was to evaluate the use of hyperspectral data to discriminate different treatments of nitrogen fertilization, and to compare different methods to obtain chlorophyll values of chlorophyll concentration based on spectral signatures of \'Mombaça\' grass. Four treatments were established with different doses of nitrogen fertilization. The spectral reflectance of leaves collected in each treatment was measured with a hyperspectral sensor, and leaf chlorophyll measurements were performed, both, on laboratory conditions. It was found that the spectral signatures of each treatments had different behaviors, mainly in the green and red-edge regions, and such differences depended on the amount of nitrogen fertilizer applied and the chlorophyll content of the leaf. The separation of the treatments was possible through the use of linear discriminant analysis, based on the spectral reflectance data obtained in each treatment. Retrieve chlorophyll concentration values from the leaf spectral reflectance was possible by the use of machine learning techniques, highlighting the support vector machine as the best alternative. The red-edge and green regions were the most influential in calculating the values of leaf chlorophyll concentration.

The influence of spectral resolution on land cover classification in Krkonoše Mts. tundra Abstract The aim of this diploma thesis was to specify the spectral resolution requirements for classification and to identify the most important spectral bands to discriminate classes of the predefined legend. Aerial hyperspectral data acquired by AisaDUAL sensor were used. The method applied for the selection of the important bands was discriminant analysis performed in IBM SPSS Statistics. The most discriminative bands were found in intervals 1500-1750 nm (beginning of SWIR), 1100- 1300 nm (longer wavelengths of NIR), 670-760 (red-edge) and 500-600 nm (green light). The classification of the selected bands was realized in ENVI 5.4 using the Support Vector Machine classifier, achieving overall accuracy of 80,54 %, Kappa coefficient 0,7755. The suitability of available satellite data for the classification of tundra vegetation in Krkonoše mountains based on spectral resolution was evaluated as well. Keywords: tundra, Krkonoše, classification, spectral resolution, class separability, discriminant analysis, hyperspectral data

Estudos com o objetivo de prever insolvência de empresas e que fazem uso de técnicas estatísticas modernas são conduzidos desde a década de 1960. Esta linha de pesquisa, que inicialmente usou técnicas univariadas, e em seguida incorporou as análises multivariadas, hoje emprega largamente técnicas que fazem uso de inteligência artificial e que necessitam uma grande capacidade de processamento computacional. Esta evolução trouxe melhorias contínuas aos resultados alcançados e hoje é possível afirmar que os demonstrativos financeiros de empresas quando analisados adequadamente são uma fonte importante de informação para a previsão de insolvência. Esta pesquisa teve como principal objetivo desenvolver e comparar modelos estatísticos usando as técnicas de Análise Discriminante Linear, Regressão Logística e Redes Neurais Artificiais a fim de investigar qual delas oferece os melhores resultados. A amostra foi composta por 40 empresas, divididas em dois grupos: o primeiro com empresas formalmente insolventes segundo os critérios da legislação brasileira, e o segundo com empresas sem tais problemas. Foram usadas inicialmente 16 variáveis para predição e empregou-se um critério de seleção de variáveis baseado nos melhores subconjuntos possíveis ao invés do stepwise. Foi tomado especial cuidado com os pré-requisitos das técnicas, sobretudo da Análise Discriminante, como normalidade e ausência de multicolinearidade das variáveis independentes. Os resultados das previsões obtidas com os modelos foram coerentes com o esperado, ou seja, a Análise Discriminante teve um desempenho inferior à Regressão Logística que também foi superada pelas Redes Neurais Artificiais. / Researches in bankruptcy prediction of companies that make use of modern statistics techniques are being held since the 1960’s. This branch of study, which initially employed univariate techniques, and then assimilated the multivariate techniques today uses artificial intelligence, a techniques that needs a great computational processing capability. This evolution brought continuing improvements to the results achieved and today is possible to say that financial statements when properly analyzed are a good source of information to the prediction of financial distress. This research aimed mainly the development of prediction models using Discriminant Analysis, Logistic Regression and Artificial Neural Networks so that they could be compared in terms of predictive capabilities. The sample consisted of 40 firms divided in 2 groups (bankrupt and non bankrupt companies) according to the Brazilian bankruptcy law. The 16 initial predictors were selected to enter the model according to the best subsets procedure in order than the stepwise procedure. Special attention was taken to accomplish the pre-requisites of the techniques, above all the Discriminant Analysis, like normality and lack of multicollinearity of the independent variables. The findings of the predictions were reasonable and according to what was expected: the Discriminant Analysis was outperformed by the Logistic Regression that was also outperformed by the Artificial Neural Networks.

Made available in DSpace on 2015-03-05T19:15:17Z (GMT). No. of bitstreams: 0 Previous issue date: 28 / Nenhuma / Este estudo, em suas abrangências teórica e prática, tem por objetivo apresentar um modelo de previsão de insolvência que retrate a realidade das micro e pequenas empresas (MPEs), fundamentado na utilização das análises discriminante e fatorial. Para atingir este objetivo realizou-se uma pesquisa qualitativa com os profissionais atuantes nos escritórios de contabilidade do Rio Grande do Sul, obtendo-se informações contábeis de 104 MPEs do período de 1995 a 2007. Para análise dos dados, foi adotado o software SPSS 10, cuja aplicação da análise fatorial reduziu o número de indica dores contábeis de 25 para 5 fatores. Para a construção da função discriminante Z, a qual permite identificar a que grupo de empresas pertence cada empresa que compõe à amostra, foi utilizada a análise discriminante. A validação do modelo foi realizada por meio do método conhecido como crossvalidation, ou seja, a subdivisão da amostra original, sendo uma para a definição do modelo e outra para a sua validação. O grau de predição do m / This study, in its theoretical and practical scope, aims to present a model for prediction of insolvency that portray the reality of micro and small enterprises (MEPs), based on the use of discriminant analysis and factor. To achieve this goal there was a qualitative research with professionals in the accounting office of Rio Grande do Sul, obtaining information accounting of 104 MEPs in the period 1995 to 2007. For data analysis, SPSS software was adopted 10, whose application of factor analysis reduced the number of pain states accounting for 25 to 5 factors. For the construction of the discriminant function Z, which allows to identify which group of companies that make each company belongs to the sample, we used the discriminant analysis. The model validation was done using the method known as crossvalidation, ie the subdivision of the original sample, one for defining the model and one for its validation. The degree of prediction of the model reached 96.15% of accuracy, representing a good index for predi

Neste trabalho se estuda a aplicação de técnicas de estatística multivariada para reconhecimento de padrões em sinais de ensaios não destrutivos (END) magnéticos, baseados no Ruído Magnético de Barkhausen (RMB). O reconhecimento de padrões pode ser feito de forma não supervisionada com a técnica multivariada de Análise de Agrupamentos, conglomerados ou Clusters que definem grupos segundo critérios de similaridade. Já para reconhecimento supervisionado a Análise Discriminante procura classificar amostras novas em grupos conhecidos, a priori, usando para este propósito uma regra de classificação criada a partir desses grupos de amostras conhecidos. Foram utilizados dois casos de detecção e classificação utilizando RMB. O RMB é um fenômeno magnético gerado por abruptas mudanças na magnetização de materiais ferromagnéticos quando submetidos a campos magnéticos variáveis. Essas mudanças estão relacionadas com a microestrutura do material, presença e distribuição de tensões elásticas (tensão e compressão). No primeiro caso de estudo procura-se identificar arames quebrados em risers, através da medição de tensão mecânica. No segundo caso procura-se classificar diferentes tratamentos térmicos em Aço AISI 420. Para a análise de integridade estrutural de risers foi feita a redução da dimensionalidade dos dados via Análise de Componentes Principais e posteriormente Análise de Agrupamentos. Já para o problema de classificação de amostras de aço foi usada a técnica de Análise Discriminante Linear de Fisher e a Quadrática. Os resultados das análises mostraram que as técnicas de Estatísticas Multivariadas proporcionam ferramentas muito adequadas para aumentar a eficiência da inspeção na área de END Magnéticos em geral e RMB em particular. / The present work deals with application of multivariate statistic techniques for pattern recognition in signals from Non-Destructive Essays (NDE), based on the Magnetic Barkhausen Noise (MBN). Pattern recognition can be done in a nonsupervised way by Cluster Analysis defining similarity criteria. On the other hand, for supervised recognition, Discriminant Analysis looks for classifying new samples in known groups, a priori, by means of classification rules created for these known sample groups. Two detection and classification cases were studied by MBN. The MBN is a magnetic phenomenon generated by sudden changes in magnetization of ferromagnetic materials, when these materials are subjected to variable magnetic fields. These changes are related to material microstructure as well as to the presence of elastic stresses (tension and compression). In the first studied case, the present study searches identifying broken wires in risers through measurements of mechanical strain. In the second case, the study classifies different thermal treatments in AISI 420 steel samples. Regarding the analysis of structural integrity of risers, firstly the reduction of data dimensionality was obtained via Analysis of Main Components and, later, Cluster Analysis was performed. Concerning the classification problem of steel samples, the Fisher Linear Discriminant Analysis and the Quadratic Analysis were used. Analysis results showed that Multivariate Statistic Techniques give rise to tools very appropriated for increasing the efficiency of inspection both in the Magnetic NDE area in general, and MBN in particular.

A Doença de Machado Joseph(DMJ) é uma doença genética autossômica dominante de início na vida adulta que afeta a coordenação motora e cursa com sintomas neurodegenerativos. É causada por uma expansão da repetição CAG no gene ATXN3. Há várias hipóteses a respeito da sua fisiopatogenia, e uma delas envolve a resposta inflamatória. O objetivo deste estudo foi descrever as concentrações séricas das citocinas em indivíduos sintomáticos, assintomáticos e compará-los com os controles saudáveis. Após a confirmação molecular dos pacientes e controles pareados por sexo e idade, os indivíduos foram convidados a participar do estudo. A idade de início e a duração da doença foram obtidas, e as escalas clínicas Scale for the Assessment and Rating of Ataxia (SARA), Neurological Examination Score for Spinocerebellar Ataxias (NESSCA), SCA Functional Index (SCAFI), and Composite Cerebellar Functional Score (CCFS), aplicadas. O soro dos indivíduos foi coletado e um painel de citocina foi realizado, incluindo a Eotaxina, GM-CSF, IFN-a, IFN-γ, IL-1b, IL-1Ra, IL-2, IL-2R, IL-4, IL- 5, IL-6, IL-7, IL-8, IL-10, IL-12, IL-13, IL-15, IL-17, IP-10, MCP-1, MIG, MIP1a, MIP1b, RANTES e O TNF-a. Entre os indivíduos sintomáticos, o painel foi repetido após 90 e 360 dias. O perfil das citocinas no baseline foi estudado por análise discriminante. Aquelas que apresentaram alterações relevantes entre os grupos tiveram seus níveis sérico reavaliados após 90 e 360 dias e estes dados foram avaliados pela equação de estimação generalizada (GEE). Sessenta e seis sintomáticos, 13 assintomáticos e 43 controles foram estudados. Quando comparados os sintomáticos e assintomáticos com seus respectivos controles saudáveis, não se observou diferenças nos padrões das citocinas. No entanto, apenas uma citocina teve destaque: os níveis séricos de Eotaxina foram significativamente mais elevados em assintomáticos (p = 0,001, ANCOVA) e entre os sintomáticos seus níveis foram menores após 360 dias do que naquelas obtidas no início do estudo (p = 0,039, GEE). A idade, a duração da doença, a expansão CAG, e as escalas NESSCA e SARA não se correlacionaram com os níveis das citocinas. O padrão relativamente benigno de citocinas em portadores sintomáticos sugere que a ativação do micróglia não seja primordial na DMJ. Entretanto, os níveis de eotaxina, um peptídeo secretado por astrócitos para repelir as células imunes circulantes, foram elevados no grupo assintomático, o que sugere que uma resposta específica destas células pode estar relacionada com a ausência de sintomas e/ou que a perda de astrócitos estaria relacionada à progressão da doença em DMJ. / Machado Joseph Disease (MJD) is an autosomal dominant genetic disease of adulthood which affects motor coordination and progresses with neurodegenerative symptoms. It is caused by an expansion of the CAG repeat at ATXN3 gene. There are several hypotheses about its pathogenesis, and one of them involves the inflammatory response. The aim of the present study is to describe the serum concentrations of a broad spectrum of cytokines in symptomatic and asymptomatic carriers of Machado Joseph disease (SCA3/MJD) CAG expansions. Molecularly confirmed carriers and controls were studied. Age at onset, disease duration, and clinical scales Scale for the Assessment and Rating of Ataxia (SARA), Neurological Examination Score for Spinocerebellar Ataxias (NESSCA), SCA Functional Index (SCAFI), and Composite Cerebellar Functional Score (CCFS) were obtained from the symptomatic carriers. Serum was obtained from all individuals and a cytokine panel consisted of eotaxin, granulocyte-macrophage colony-stimulating factor (GM-CSF), interferon (IFN)-α, IFN-γ, interleukin (IL)-1β, IL-1RA, IL-2, IL-2R, IL-4, IL-5, IL-6, IL-7, IL-8, IL-10, IL- 12, IL-13, IL-15, IL-17, interferon gamma-induced protein (IP)-10, monocyte chemoattractant protein (MCP)-1, monokine induced by gamma interferon (MIG), macrophage inflammatory protein (MIP)-a, MIP-b, regulated on activation, normal T cell expressed and secreted (RANTES) and tumor necrosis factor (TNF)-α was analyzed. In a subgroup of symptomatic carriers, the cytokine panel was repeated after 90 and 360 days. Cytokine distribution among groups was studied by discriminant analysis; changes in serum levels after 90 and 360 days were studied by generalized estimation equation. Sixty-six symptomatic carriers, 13 asymptomatic carriers, and 43 controls were studied. No differences in cytokine patterns were found between controls and carriers of the CAG expansions or between controls and symptomatic carriers only. In contrast, eotaxin concentrations were significantly higher in asymptomatic than in symptomatic carriers or in controls (p = 0.001, ANCOVA). Eotaxin did not correlate with age, disease duration, CAG expansion, NESSCA score, and SARA score. Among symptomatic carriers, eotaxin dropped after 360 days (p = 0.039, GEE). SCA3/ MJD patients presented a benign pattern of serum cytokines. In contrast, levels of eotaxin, a peptide secreted by astrocytes, were elevated in the asymptomatic carriers, suggesting that a specific response of these cells can be related to symptom progression, in SCA3/MJD.

Finite mixture modeling is a useful statistical technique for clustering individuals based on patterns of responses. The fundamental idea of the mixture modeling approach is to assume there are latent clusters of individuals in the population which each generate their own distinct distribution of observations (multivariate or univariate) which are then mixed up together in the full population. Hence, the name mixture comes from the fact that what we observe is a mixture of distributions. The goal of this model-based clustering technique is to identify what the mixture of distributions is so that, given a particular response pattern, individuals can be clustered accordingly. Commonly, finite mixture models, as well as the special case of latent class analysis, are used on data that inherently involve repeated measures. The purpose of this dissertation is to extend the finite mixture model to allow for the number of repeated measures to be incorporated and contribute to the clustering of individuals rather than measures. The dimension of the repeated measures or simply the count of responses is assumed to follow a truncated Poisson distribution and this information can be incorporated into what we call a dimension informative finite mixture model (DIMM). The outline of this dissertation is as follows. Paper 1 is entitled, "Dimension Informative Mixture Modeling (DIMM) for questionnaire data with an informative number of repeated measures." This paper describes the type of data structures considered and introduces the dimension informative mixture model (DIMM). A simulation study is performed to examine how well the DIMM fits the known specified truth. In the first scenario, we specify a mixture of three univariate normal distributions with different means and similar variances with different and similar counts of repeated measurements. We found that the DIMM predicts the true underlying class membership better than the traditional finite mixture model using a predicted value metric score. In the second scenario, we specify a mixture of two univariate normal distributions with the same means and variances with different and similar counts of repeated measurements. We found that that the count-informative finite mixture model predicts the truth much better than the non-informative finite mixture model. Paper 2 is entitled, "Patterns of Physical Activity in the Northern Manhattan Study (NOMAS) Using Multivariate Finite Mixture Modeling (MFMM)." This is a study that applies a multivariate finite mixture modeling approach to examining and elucidating underlying latent clusters of different physical activity profiles based on four dimensions: total frequency of activities, average duration per activity, total energy expenditure and the total count of the number of different activities conducted. We found a five cluster solution to describe the complex patterns of physical activity levels, as measured by fifteen different physical activity items, among a US based elderly cohort. Adding in a class of individuals who were not doing any physical activity, the labels of these six clusters are: no exercise, very inactive, somewhat inactive, slightly under guidelines, meet guidelines and above guidelines. This methodology improves upon previous work which utilized only the total metabolic equivalent (a proxy of energy expenditure) to classify individuals into inactive, active and highly active. Paper 3 is entitled, "Complex Drug Use Patterns and Associated HIV Transmission Risk Behaviors in an Internet Sample of US Men Who Have Sex With Men." This is a study that applies the count-informative information into a latent class analysis on nineteen binary drug items of drugs consumed within the past year before a sexual encounter. In addition to the individual drugs used, the mixture model incorporated a count of the total number of drugs used. We found a six class solution: low drug use, some recreational drug use, nitrite inhalants (poppers) with prescription erectile dysfunction (ED) drug use, poppers with prescription/non-prescription ED drug use and high polydrug use. Compared to participants in the low drug use class, participants in the highest drug use class were 5.5 times more likely to report unprotected anal intercourse (UAI) in their last sexual encounter and approximately 4 times more likely to report a new sexually transmitted infection (STI) in the past year. Younger men were also less likely to report UAI than older men but more likely to report an STI.

The concern of this thesis is to discuss classification of different objects based on longitudinal observations. In the first instance the reader is introduced to a linear mixed-effects model which is useful for longitudinal data modeling. Description of discriminant analysis methods follows. These methods ares usually used for classification based on longitudinal observations. Individual methods are introduced in the theoretic aspect. Random effects approach is generalized to continuous time. Subsequently the methods and features of the linear mixed-effects model are applied to real data. Finally features of the methods are studied with help of simulations.

The resource-based view of the firm (RBV) is one the most important areas of research content to emerge in the field of strategic management in the last 15 years. The RBV is prescriptive. That is, the RBV prescribes that competitive advantage stems from those resources that are valuable, rare, inimitable, and nonsubstitutable (VRIN). With rare exception, resources that meet the VRIN criteria are widely purported to be intangible in nature. From a research perspective, the RBV stream tends to be dominated by conceptual discussions and advancements. However, empirical tests of the core premises, or the main prescription, of the theory are argued to be very limited in quantity. To add to the body of empirical research that seeks to verify the main prescription of the RBV, this research undertakes a new and different level of analysis, one that has not been previously tested. Given that firms compete with both tangible and intangible resources, the present study is interested in determining if, as the RBV implicitly prescribes, resources that are intangible in nature are more important determinants of firm success than tangible resources. Although the research question is basic and fundamental, it has rarely been appropriately or adequately tested within the RBV stream, as is demonstrated by this thesis. To carry out the research, this study offers a conceptual model of the firm’s resource pool that includes tangible assets (financial and physical assets), intangible assets (intellectual property assets, organizational assets, reputational assets), and capabilities. A series of hypotheses are posited to explore the proposition that intangible resources contribute more greatly to firm success, on the dimensions of sales turnover, market share, and profitability, than tangible resources. / A field survey, administered to 2000 manufacturing and services businesses operating in Australia, is used to gather the data. Of the 2000 surveys sent, the hypotheses are empirically tested using multiple hierarchical regression analysis on a final sample of 291 firms. Control variables include firm age and Porter’s five forces of industry structure. Based on the results, verification of the RBV’s main prescription can not be supported unequivocally. Intellectual property assets, for example, do not have a statistically significant association with firm success, after accounting for the effects of tangible resources and the control variables. Organizational assets, however, not only explain additionally significant variation in firm success, after accounting for the effects of tangible resources and the control variables, but make among the greatest, unique contribution to firm success based on the size of the beta coefficients. Reputational assets offer additional explanatory power to predicting firm success after accounting for the effects of tangible assets and the control variables, but only with respect to one measure of firm success does its beta coefficient make a larger, unique contribution than financial assets. Lastly, contrary to theory, capabilities are not the single most important determinant of firm success, after accounting for the effects of intangible assets, and tangible and intangible assets, in two separate hierarchical regression equations. This finding is surprising and explanations are provided. Overall, the study raises some questions with respect to just which resources are the most important determinants of a firm’s market and financial success and offers a fruitful avenue for further research.