MULTIVARIATE ANALYSIS

Author: ANDRADES CALDITO LIDIA.
Year: 2002.
University: EXTREMADURA [www.unex.es].
Place of preparation: FACULTAD CIENCIAS ECONÓMICAS Y EMPRESARIALES.

Summary: Analysis of the purchasing behavior of the consumiodres from the perspective of global models of consumer behavior (Alonso, 2000). Contrastación empirical hypotheses raised using cluster models of calse latent and structural equation models.

Author: CONTATO CAROL M. LUCRECIA.
Year: 2003.
University: POLITÉCNICA DE MADRID [www.upm.es].
Place of defense: ESCUELA TECNICA SUPERIOR DE INGENIEROS DE MONTES.
Place of preparation: ESCUELA TECNICA SUPERIOR DE INGENIEROS DE MONTES.

Summary: Summary: The integrity psico -- physical men in the cities has been threatened since the implosion of growth thereof, produced mainly from the second half of the century XX.El development sudden large nuclei, has resulted in the hacianamiento. The lack of viviendad, inadequate services and above all, ocntaminación all formas.atmosférica, groundwater, noise, increased temperature, along with other more subtle, less visible, but no less impotantes: stress citizen and the loss of their cultural and historical identity. The presence of trees and landscapes created by them in the cities, has been partly offset esso effects negativos.Las squares, parks, the trees in the sidewalks, from platabandas in the avenue, private gardens, etc.. They do not carry themselves that the other purpose of reintegrating nature in the city. But the growth of cities and increasing their population densities and building has led to the availability of free land to accommodate trees, is increasingly scarce, hence its incorporation and maintenance, increasingly costly and sacrificed. In these circumstances, when valuation of the trees or knowledge about its value, it becomes necessary to put them on equal footing with the other elements that make up the landscape urbano.Encontrar a value for losárboles of the city without means place dudad, place them in a position of prominence and respect within society. We present in this paper a procedure for calculating parametric value of urban trees, based based on the capitalization of expenses that become the management Municipal pair produce, planting, protecting and preservarlos.Se has considered especially, a baneficio derived from its presence: coverage of the trees in the city, tomándosela as fundamental variable and greater weight in the rating. The construction of indexes that increase or decrease their value, comtempla groups of species for postage, característivas intrísecas, growth rates, the attractiveness of flowering, quality of the shade-producing and most importantly, the location of the plant within the ejido urban, giving greater copies located in neighborhoods of low coverage.

Author: CORTES SAUD ANGELA DEL CARMEN.
Year: 2004.
University: SALAMANCA [www.usal.es].
Place of defense: DEPARTAMENTO DE ESTADISTICA.
Place of preparation: DEPARTAMENTO DE ESTADISTICA.

Summary: In the analysis of three-way tables are often available for additional information or additional study is necessary to incorporate the structure of this type of multivariate data. To achieve this goal we introduced the Principal Component Analysis (ACPR) Restricted to three-way data, which makes it possible to incorporate additional information, each of the different ways of settlement data, the representative structure of these. In a first phase, the ACPR specifies a general model which assumes significance defining matrices spaces on projection sets, and thereby matrices metric associated with these spaces. In a second phase, for the analysis of the terms of the model or combinations of them will be necessary to use the concept of Generalized Singular value decomposition in (DVSG), extended to three tracks. The Canon Correspondence Analysis of two tracks (PCA: Ter BRAAK, 1986) was established in the framework of analysis Vegetation Environmental Sciences, today its application has spread to many areas such as studies of Public Health, Marketing Impact Environmental Biosanitarios, etc.. In many research designs, both in ecology and other areas of knowledge, measurements were repeated on different occasions, namely data can be sorted in a three-way settlement. Our proposal, the PCA three-way, comes from extending the Principal Component Analysis Restricted two routes (ACPR: TAKANE and HUNTER, 2001), which is a versatile technique for analyzing the structure of multivariate data and is widely applicable. In the analysis of three-way tables are often available for additional information or additional study is necessary to incorporate the structure of this type of multivariate data. To achieve this goal we introduced the Principal Component Analysis (ACPR) Restricted to three-way data, which makes it possible to incorporate additional information, each of the different ways of settlement data, the representative structure of these. In a first phase, the ACPR specifies a general model which assumes significance defining matrices spaces on projection sets, and thereby matrices metric associated with these spaces. In a second phase, for the analysis of the terms of the model or combinations of them will be necessary to use the concept of Generalized Singular value decomposition in (DVSG), extended to three tracks. The Canon Correspondence Analysis of two tracks (PCA: Ter BRAAK, 1986) was established in the framework of analysis Vegetation Environmental Sciences, today its application has spread to many areas such as studies of Public Health, Marketing Impact Environmental Biosanitarios, etc.. In many research designs, both in ecology and other areas of knowledge, measurements were repeated on different occasions, namely data can be sorted in a three-way settlement. Our proposal, the PCA three-way, comes from extending the Principal Component Analysis Restricted two routes (ACPR: TAKANE and HUNTER, 2001), which is a versatile technique for analyzing the structure of multivariate data and is widely applicable.

Author: DAUNIS I ESTADELLA JOSEP.
Year: 2004.
University: POLITÉCNICA DE CATALUÑA [www.upc.edu].
Place of defense: SALA DE JUNTES. FME.
Place of preparation: EDIFICI A0 - OMEGA Campus NORD.

Author: JUAREZ ALONSO CARLOS ALBERTO.
Year: 2004.
University: POLITÉCNICA DE CATALUÑA [www.upc.edu].
Place of defense: SALA TELEENSENYAMENT DE L'ETSETB (B3).
Place of preparation: EDIFICI A0 - OMEGA Campus NORD.

Summary: The attitudes, knowledge and actions often are based on samples. Some based their conclusions on small samples and rarely take into account the magnitude of which is unknown. It is generally lacks the resources to explore more of a part of the problem of interest that might increase our knowledge. Some reasons for the use of sampling techniques are: reduced cost, increased speed, greater focus or perspective and greater accuracy. The merging of data is emerging as an alternative to the single source of data compared to the need to get as much information as possible at the lowest cost. It aims to combine data from different sources in order to have all the information in a single file, even if artificial, with all the variables of interest. What better use of existing information in a file to reconstruct the missing information in another file. It is a statistical estimation of missing data. It is a means to limit the collection of data, reconstructing missing information. It is not a problem of statistical analysis with missing data are considered in which the mechanisms leading to the absence of data. In the case of data fusion are presented complete block of data missing, in general, independent samples. The literature review has enabled analysis of the problem of missing data analysis as an approximation to the issue of data fusion, considering that the merger of data is a special case. In this case, there is talk of missing data blocks and on many occasions, missing data by design. The goal of data fusion is thus, get a single file that can be analyzed later with data mining tools. The idea is to estimate the values of the variables not informed (missing values) from a block of variables correlated with the reported block variables to reconstitute (common variables). It is important to note that this is not generating new information. Part of this work is devoted to the definition and alignment of data fusion concept. We present some approaches to the treatment of missing data. Have been revised, so not exhaustive, some techniques related to the subject and show applications of data fusion-related tools. Many techniques exist to deal with the reconstruction of missing data. In this paper, the focus has been on the technique of charging hot deck. With this technique are replaced individual values drawn from similar units observed. It is based on the replacement of the missing information in all individuals not reported (with missing data), which are called receptors, using information from individuals reported more similar, called donors. It is a common practice that involves schemes developed for the selection of units similar to the complaint. Featured are some non - parametric procedures of discrimination for individuals seeking closer (similar). Different algorithms designed for this. There has been a review of some of these search methods. This work is based on the algorithm Fukunaga / Narendra. It shows the system components. This paper presents the rationale and the foundations that have been followed for the development and implementation of the same. Some proposals have been established for the non - parametric measurement of the quality of the merger. The experimental and test different proposal has been made in two parts. In one has made a random selection of individuals to form a joint donor and a set of receivers. In the second part, the selection had been made from a specific characteristic in a variable.

Author: CASANOVA GURRERA M. DESAMPARADOS.
Year: 2004.
University: BARCELONA [www.ub.es].
Place of defense: FACULTAD DE BIOLOGÍA.
Place of preparation: FACULTAD DE BIOLOGÍA.

Author: HERNANDEZ GONZALEZ SERGIO.
Year: 2004.
University: SALAMANCA [www.usal.es].
Place of defense: SALON DE GRADOS, EDIFICIO FES.
Place of preparation: DEPARTAMENTO DE ESTADISTICA.

Summary: Generally attention to the potential bias implicit in an investigation, since it is well known that bias undertake the external validity of the studies, but despite that biases are kept under control, and data are truly representative the results of the statistical treatment can be distorted for many reasons. This does not happen in Univariate and bivariate analysis, but also in the multivariate analysis, and especially in the technical factor. This has been the starting point of our investigation, because our goal is focused on Biplot analysis proposed by Gabriel in 1971 and the question that prompted this investigation is: How  affects the presence of discordant data to the structure factor associated with Biplot analysis? The main objective of Biplot, is to provide the approximate graphic reproduction of the data, with a quality of representation that can, thanks to their geometric properties, capture visually the interrelationships between sets of individuals and variables, as well as the relationships between elements each of these sets. Its theoretical foundation is based on the approximation of a certain range of data matrix, by a low-ranking, through the decomposition in Singular Values, and then make a factoring markers in rows and columns. Biplot Figures are based on scalar and products are interpreted in terms of projections, angles and distances to the center of gravity and / or between the markers, so hopefully a distorting effect on the structure of covariance, motivated by the presence of data discordant (known in most literature as outliers) with the general pattern, which may be due to a procedural error, an observation that occurs as a result of an extraordinary event, such as comments that are outside the range of ordinary values of each variable, or as comments still remain within the expected range of each variable are "unique" in its combination of values between variables. At the time of this work plan are only two references, DEIGLE (1989) and DEIGLE and RIVEST (1992), which raised the robustness of the methods Biplot compared to the presence of outliers. This was the core of our investigation.

Author: SÁIZ ABAJO Ma. JOSÉ.
Year: 2004.
University: LA RIOJA [www.unirioja.es].
Place of defense: UNIVERSIDAD DE LA RIOJA.
Place of preparation: UNIVERSIDAD DE LA RIOJA.

Summary: Fraud is a growing phenomenon in recent times due to the opening of international markets and global competition. The aim of the doctoral thesis which is entitled "Characterization, classification and detection of fraud in vinegar through multivariate techniques" is to limit or eliminate these risks of counterfeiting in the wine vinegar. In the case of industry vinagrera in Spain, control and detection of fraud is the responsibility of the Spanish Association of Vinegar. However, the analytical tools contained in the Technical Regulations Sanitary Spanish are insufficient to detect fraud more common in the vinegar with some reasonable limits of detection. The increase in these scams can lead to irreversible economic consequences for the industry honored the profession that complies with legislation. Fraud in vinegar can be reduced to two situations: 1) the addition of acetic acid in vinegar to non-agricultural origin to comply with the legal specifications, and 2) mixing different proportions of vinegar wine vinegar and alcohol and the subsequent sale of the resulting product under the name of genuine wine vinegar. The first of the adulteration may be cited currently detected by the development of analytical methods based on nuclear magnetic resonance. However, the second of the aforementioned frauds is still a problem unsolved. This thesis presents the near-infrared spectroscopy (NIR) as a technique for rapid and useful classification and characterization of vinegar, and for the detection of the second type of fraud. Studies conducted show that the technique NIR is appropriate to characterize and quantify up to a total of fifteen physical and chemical parameters related to the quality of vinegar, in addition to the measurement of percentage of adulteration in the case of vinegars resulting from the mixing of raw materials. To complete the study and make it applicable to the industry, proposes a number of strategies for robust NIR calibration. The conclusions drawn from this dissertation provide analytical tools and results that can be applied in industry vinagrera to control the process of preparing the vinegar, the same quality and detect possible adulteration.

Author: ESTEBAN DÍEZ ISABEL.
Year: 2004.
University: LA RIOJA [www.unirioja.es].
Place of defense: UNIVERSIDAD DE LA RIOJA.
Place of preparation: UNIVERSIDAD DE LA RIOJA.

Summary: The applicability of the real near-infrared spectroscopy (NIRS) in the control and monitoring processes across multiple line industrial applications depends on determining the robustness and reliability associated with model calibration and classification multivariate built, as these serve as models foundation to carry out quantitative or qualitative analysis of critical parameters of quality. For this reason, this dissertation focuses on the optimization of the quality of model calibration and classification constructed on the basis of samples NIR steps involved in various industrial manufacturing processes, thanks both to the implementation of some convenient tools quimiométricas existing as the development of new methodologies for proper treatment and modeling of data. This dissertation is presented as a compendium of scientific papers, following the rules of the University of La Rioja. The ten articles submitted have been grouped into three distinct blocks perfectly, depending on the type of practical application and the specific objectives that address.

Author: MARTIN DE DIEGO ISAAC.
Year: 2005.
University: CARLOS III DE MADRID [www.uc3m.es].
Place of defense: ESCUELA POLITECNICA SUPERIOR.
Place of preparation: UNIVERSIDAD CARLOS III DE MADRID.

Summary: Support Vector Machines are methods for solving problems of classification and regression. It is a technique based on the translation of data into a space larger than the original using a core function and the search for a hiperplano separator in the new space. Choosing the core function which defines the new space is an issue. In this thesis we deal with the combination of a collection of kernel functions using labels class. Schemes combination we propose is based on the quantification of the difference of information between cores and on the concept of functional combination of nuclei. Sometimes the matrix constructed loses certain desirable properties of any parent nucleus. We propose methods for transforming parent not verify these properties. Our techniques are based on the approximation of an array to another kernel which is not without first losing the property if semidefinida positive. We propose the method Kernel Procrustes, a scheme combining core functions homogeneous alternative to the average of the nuclei. We do extensive work on experimental data collections artificial and real.

Author: ALFARO CORTES ESTEBAN.
Year: 2005.
University: CASTILLA-LA MANCHA [www.uclm.es].
Place of defense: FACULTAD DE CIENCIAS ECONÓMICAS Y EMPRESARIALES DE ALBACETE.
Place of preparation: FACUTLAD DE CIENCIAS ECONÓMICAS Y EMPRESARIALES DE ALBACETE.

Summary: The work is structured in three parts: Part I covers chapter to chapter IV. After presenting some general aspects of the problems of classification, discusses some of the more individual grading methods used at present, highlighting its main advantages and disadvantages. The second part (Chapters V to VII) discusses some aspects of the behavior and properties of the individual classifiers. Specifically, it raises the difficulties that may arise due to the use of the individual classifiers, such as the accuracy and stability of the same. Then, in Chapter VI deals with the study of the combination of classifiers paying special attention to boosting method. In addition, it includes a taxonomy of the methods of combination and also introduces the method bagging and the forest random. Finally, we study the early algorithms that have resulted in the development of post-boosting method. It also highlights some of the changes that have been proposed algorithm Adaboost, including those that serve to address the existence of two other classes and, to finish, he examines what should be the appropriate size of the trees used in combination. The third part (chapters 8-10) provides an overview of the prediction business judgment, their background and current status. In addition, it will develop a list of financial ratios that have proved most useful for the prediction of failure. It also carries out a description of the developments that have followed in Spain enterprises failed. Chapter IV focuses on the practical application. After collecting some brief theoretical considerations on the treatment of information, an analysis of exploratory data. In addition fourteen financial ratios, which are used by other three variables rarer who attempt to collect the size of the company, the activity for which it is dedicated, and the legal form it presents. It collates boosting with the method of classification trees, both for the case dichotomous, as when three types. Below is a comparison, slightly less detailed, with five other classification methods. Thereafter, it examines the ability of the models to predict the previously established business failure increases when the distance to the temporary period in which the ruling becomes effective. Finally, it concludes that boopsting improves outcomes of the individual classification trees. Among the major contributions of this work include the use of a novel technique, the method and boosting the consideration of a concept of business failure broader usual.

Author: GUBERN FORTUNY MARCEL.
Year: 2005.
University: POLITÉCNICA DE CATALUÑA [www.upc.edu].
Place of defense: seminari 1, planta 7.ETSEIB.
Place of preparation: ETSEIB, EDIFICI H Campus SUD.

Author: BASSO LUIZ CARLOS.
Year: 2005.
University: SALAMANCA [www.usal.es].
Place of defense: EDIFICIO FES.
Place of preparation: DEPARTAMENTO DE ESTADISTICA.

Summary: In this paper we address the joint analysis of several arrays of data and the possible ways of representation, with a special interest in the study of individuals. They studied three techniques proposed in the literature: the method Statis, method of Krzanowski and analysis triádico. These techniques in its original formulation, are based on the GPA, and do not provide simultaneous representation of individuals and variables and no representation of the evolution of individuals. For the first two techniques, the school has proposed salmantina versions biplot. Following the work in that direction, we propose a version triádico biplot for analysis, which we call Biplot Triádico, which enables the two types of representation referrals. We present an initial study of the three methods, biplot versions available for the first two and the theoretical development of the biplot Triádico. We also propose several measures of adjustment. To facilitate the implementation of the theoretical results, we have developed a program that we Matlab entitled BIPTRIAD. The program also enables the comparison between the spaces commitment biplot of Krzanowski and biplot Triádico. Finally, we present the application of theory to a real data set of Foreign Direct Investment in Europe in Latin America using the BIPTRIAD, which demonstrated the theoretical possibilities of the method and the possibilities of BIPTRIAD both to implement and to allow comparison empirical between biplot of Krzanowski and Triádico for each particular data set.

Author: RIOS MORENO ROXANA IVETH.
Year: 2005.
University: SALAMANCA [www.usal.es].
Place of defense: SALA DE GRADOS DE LA FACULTAD DE MEDICINA.
Place of preparation: DEPARTAMENTO DE ESTADISTICA.

Author: MABEL WILLEMS PRISCILA.
Year: 2005.
University: SALAMANCA [www.usal.es].
Place of defense: DEPARTAMENTNO DE ESTADISTICA.
Place of preparation: DEPARTAMEMTO DE ESTADISTICA.

Summary: The analysis of ecological information falls within the gradient analysis, which aims to study the spatial patterns of species, using information of the species, their characteristics and environment variables. The underlying model that relates the species with environmental gradients is generally non-linear, as commonly assumed the gaussian model. A multivariate level, this kind of study can be done through methods of management or restricted management. In this paper we consider two types of methods as complementary strategies, and terms of a general model of partition of the total variability of the species information. A comparison of their results to assess, empirically, the validity of the assumptions involved in the external constraints. This was the extension of the model Correspondence Analysis No. Symmetrical towards a restricted management, in a context of Direct Analysis of gradient, according to the species in response to the location. This technique, which we call Canon Correspondence Analysis No. Symmetrical (CNCA), work from two approaches: analysis of projected values (perfiles-filas projected), and analysis of a pivot table, from which estimators can obtain external variables representing relations marginal. Areas of management are restricted to be linear combinations of the environmental variables, based on information from the most abundant species. Get estimators markers locations, species and environmental variables. It presents indicators goodness of fit, and describes indicators contributions elements and factors as elements of the diagnostic interpretation of the representations factorials. The latter are interpreted through biplots projections, submitted an application to real data of mallines and steppe in Patagonia, Argentina.