FREE AND NON - PARAMETRIC DISTRIBUTION METHODS

Author: Jácome Pumar Ma. Amalia.
Year: 2004.
University: A CORUÑA [www.udc.es].
Place of defense: Facultad de Informática.
Place of preparation: Facultad de Informática.

Summary: This report deals with the study of variables measuring times of life, namely that a certain event runs from baseline to end an event. However, individuals are often those who are not able to record the final moment of the event. Of these individuals, called data censored by the right, only knows that his time real life is superior to that observed. For example, in medicine, a patient who survives at the end of the clinical trial or dying from another cause other illness in the study. The stimulator consists of the distribution function in this context is the estimator Kaplan-Meier (1958). This report presents a new method of estigmación, for the functions of distribution and density, which lies in the previous estimate, through non - parametric procedures, the role of conditional probability is not censorship. This preliminary step, called presuavización leads to a more efficient use of the information provided in the data. Issues related to the properties asintóticas (representation, consistency and normality) estimator presuavizado of the distribution function and shows their efficiency, in a simulation study, regarding the Kaplan-Meier estimator. The same properties asintóticas are shown for the estimator presuavizado of density. In addition, we obtain a representation of asymptotic mean square error integrated (MISE) and the best that it minimized windows. Based on the representation of the windows, a plug-type selector and test its consistency probability. The behavior of this practice, along with other selectors based remuestreos bootstrap has been analyzed in a simulation study. It examines also the effect on the estimate presuavizada when using two different estimators of the conditional probability of non-censorship. Specifically, have compared the estimators Nadaraya-Watson and local linear. Finally, it illustrates the behavior of both estimators presuavizados with an application to data on the heart transplant program at Stanford.

Author: PRATS MONTALBÁN JOSÉ MANUEL.
Year: 2004.
University: POLITÉCNICA DE VALENCIA [www.upv.es].
Place of defense: Dep. Estadistica e Investigacion Operativa.
Place of preparation: Universidad Politécnica de Valencia.

Summary: 1. Introduction The lower the acquisition of color images in industrial processes has led to the ability to control them from the automated visual inspection of the first. Thus, taking a color image that characterizes the process in normal operation, it is possible to create a form that accumulate as much relevant information as possible, and from the same whether new images to be consistent with it, or not. One way to enrich the information provided by these images is to include type of textual information, because it can help characterize the image used to create the model as well as to determine the existence of new features or flaws in the new images. 2. Data structure Arrays of dimension 4 forming such images are actually three different types of modes or addresses of natural variation: the sample information, as defined by the two-dimensional structure of the pixels, which are the objects that are to analyze, and are deployed along the first direction; information in a textual, composed of pixels which is adjacent to the sample and forming the window neighborhood and the character of spectral information, formed by the various channels color through which contains information of this type. Thus obiene an internal structure 3-way type OVV (Object, variable, variable). 3. Multivariate Analysis of images. The multivariate analysis of these images 3-way can be carried out using methodologies deployed in the matrix of type Unfold-PCA or Unfold-PLS or directly from the models N-way such as PARAFAC, Tucker 3 or No. - PLS. 4. Approach classification and detection of defects. Once characterized the / a picture / is "pattern," you can determine the adequacy of the new images / s first / s from

Author: CHACÓN DURÁN JOSÉ ENRIQUE.
Year: 2004.
University: EXTREMADURA [www.unex.es].
Place of defense: FACULTAD DE CIENCIAS.
Place of preparation: FACULTAD DE CIENCIAS.

Summary: The report examines the problem of estimating densities in Mathematical Statistics. In a very general framework sets out the relationship between the concept of sufficiency and the estimated partial densities and also to test the hypothesis that the density is in some space Lp is not distinguishable, in the sense Hoeffding and Wolfowitz. For the specific case of a density estimate of actual univariate explores the kernel estimator, with particular emphasis on the problem of choice of an optimum bandwidth. Under conditions are very restrictive test the existence of the optimal bandwidth, as well as their properties asistólicas and describes the succession of bandwidths optimal order until the third key word when using a core of any order k. It describes the methodology bobtstrap, representing statistics structures suitable for the study of estimators bootstrap, both for fixed sample size to the asymptotic case, and this method is applied to the problem of automatic choice of bandwidth pilot type local and studying their properties asintóticas through theorems distribution limit of relative error regarding dela succession optimal bandwidths, and their behavior for finite sample sizes, through a comparative study by simulation.

Author: PÉREZ MARÍN ANA MARÍA.
Year: 2005.
University: BARCELONA [www.ub.es].
Place of defense: FACULTAD DE CIENCIAS ECONÓMICAS Y EMPRESARIALES.
Place of preparation: FACULTAD DE CIENCIAS ECONÓMICAS Y EMPRESARIALES.

Summary: The central theme of this thesis is the loyalty of customers and risk management business insurers. The thesis discusses the duration of insured as a client of the insurance company through the use of a -- proposed for this purpose and called "Naive Local constant." It analyzes data on the cancellation of an insurance policy provided by Danish and draws conclusions sobrelos factors associated with a shorter duration of the life of the insured within the company and on the risk management business in the same.