Inferencia estadística en procesos puntuales sobre grafos lineales
Loading...
Identifiers
Publication date
Authors
Advisors
Tutors
Editors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Este trabajo constituye un recorrido que, desde el nivel del alumnado del Grado en Matemáticas, permite llegar a abordar problemas de inferencia no paramétrica sobre la función de intensidad de procesos puntuales definidos sobre grafos lineales. Para ello, hemos establecido una estructura en capítulos que ha de ser entendida como una consecución de peldaños de complejidad ascendente, en los que, de manera gradual, se van presentando los conceptos y herramientas necesarias hasta llegar al desarrollo del problema final (y objetivo de este trabajo). Comenzamos introduciendo una serie de conceptos esenciales de la Estadística y la Teoría de la Probabilidad que se necesitan y manejan a lo largo de todo el trabajo. En el Capítulo 1 nos centramos en un problema clásico y bien conocido que es la estimación de la función de densidad, focalizándonos en las técnicas no paramétricas, y en particular en los métodos núcleo. Este capítulo no ha de entenderse una mera introducción, ya que es de esencial importancia en el estudio de los procesos puntuales por la íntima relación existente entre las funciones de densidad y las funciones de intensidad. En el Capítulo 2, presentamos los procesos puntuales en el plano euclídeo y sus modelos esenciales, como los procesos de Poisson. Una vez que se manejan estos conceptos sobre el plano euclídeo, en el Capítulo 3 se incrementa un grado el nivel de complejidad presentándolos sobre grafos lineales, en donde ya no disfrutamos de ventajas como la existencia de la métrica euclídea o del concepto de gradiente. De nuevo, focalizamos nuestro estudio en los procesos de Poisson, estudiando en profundidad diversas técnicas de estimación no paramétrica de la función de intensidad, así como la cuestión clave de los selectores del parámetro ventana. Uno de los problemas más interesantes y estudiados en la literatura estadística es el de comparación de dos (o más) poblaciones. El Capítulo 4 se dedica íntegramente a la presentación de este problema en el marco de los procesos puntuales en grafos lineales. Además, se aportan soluciones innovadoras que consisten en tres test estadísticos que permiten concluir si, dados dos patrones de puntos sobre un mismo grafo lineal, provienen de procesos con funciones de intensidad proporcionales, es decir, con una densidad espacial común que se traduce en una misma estructura espacial (en términos de propiedades de primer orden).
En el Capítulo 5 se lleva a cabo un exhaustivo estudio de simulación con varios escenarios en los que se comprueba la calidad y buen comportamiento de los métodos de contraste propuestos en el capítulo anterior. Para ello se presentan tanto resultados sobre el ajuste de cada contraste en distintos tipos de grafos lineales, así como de la potencia de los mismos.
Para finalizar el trabajo, se presenta una aplicación de parte de los métodos descritos para
el contraste de dos poblaciones sobre un conjunto de datos reales de accidentes de tráfico en la ciudad de Río de Janeiro (Brasil) entre 2019 y 2022. Además, en el apéndice de este trabajo puede encontrarse el código empleado, con garantías de reproducibilidad. Cabe decir que no se incluye la base de datos reales por cuestiones de confidencialidad
This work constitutes a journey which, starting at an undergraduate level, will allow the addressing of non-parametric problems dealing with point processes on linear network's intensity function. To do so, a chapter structure has been established, which must be understood as a series of complexity-ascending steps. Gradually, the concepts and techniques required to fullyunderstand the final problem (and aim of this work) will be presented. Firstly, some Statistics and Probability Theory core concepts will be introduced, as they will be used all through our discussion. Chapter 1 is centered on the well-known density function's estimation problem, focusing on non-parametric techniques, particularly kernel methods. This chapter must not be understood as a mere introduction, as it is of utmost importance when studying point processes due to the intimate relation exsiting between density and intensity functions. In Chapter 2 point processes on the euclidean plane, and their essential models such as Poisson processes, are presented. Once familiarised with these concepts on the euclidean plane, a step forward is taken in Chapter 3, introducing them on linear networks, where the euclidean distance or concepts such as gradient are no longer available. Once again, our study focuses on Poisson processes, studying in great length diverse intensity function's non-parametric estimators, plus the key issue of bandwidth selection. One of the most interesting, and studied, problems in statistical literature is that of comparing two (or more) populations. Chapter 4 is dedicated integrally to this problem's discussion in the point processes on linear network's framework. Furthermore, innovative solutions consisting of three statistical test are presented. Given two point patterns on the same linear network these test allow to determine whether those patterns are realizations of two point processes with proportional intensity functions; meaning they share density function, and therefore have the same spatial structure (in terms of first-order properties). In Chapter 5 an exhaustive simulation study is performed, analysing in various scenarios the quality and well-behaving of the contrast methods proposed in the previous chapter. In order to do so, results of both level and power of those test are presented in different linear networks. Finally, some of the proposed methods have been applied so as to contrast two populations, based on a real-life dataset of trafic accidents in Río de Janeiro (Brasil) from 2019 up to 2022. Moreover, in this work's appendix the code which has been used, with guaranteed reproducibility, can be found. It is worth mentioning that the database has not been included due to condidentiality issues
This work constitutes a journey which, starting at an undergraduate level, will allow the addressing of non-parametric problems dealing with point processes on linear network's intensity function. To do so, a chapter structure has been established, which must be understood as a series of complexity-ascending steps. Gradually, the concepts and techniques required to fullyunderstand the final problem (and aim of this work) will be presented. Firstly, some Statistics and Probability Theory core concepts will be introduced, as they will be used all through our discussion. Chapter 1 is centered on the well-known density function's estimation problem, focusing on non-parametric techniques, particularly kernel methods. This chapter must not be understood as a mere introduction, as it is of utmost importance when studying point processes due to the intimate relation exsiting between density and intensity functions. In Chapter 2 point processes on the euclidean plane, and their essential models such as Poisson processes, are presented. Once familiarised with these concepts on the euclidean plane, a step forward is taken in Chapter 3, introducing them on linear networks, where the euclidean distance or concepts such as gradient are no longer available. Once again, our study focuses on Poisson processes, studying in great length diverse intensity function's non-parametric estimators, plus the key issue of bandwidth selection. One of the most interesting, and studied, problems in statistical literature is that of comparing two (or more) populations. Chapter 4 is dedicated integrally to this problem's discussion in the point processes on linear network's framework. Furthermore, innovative solutions consisting of three statistical test are presented. Given two point patterns on the same linear network these test allow to determine whether those patterns are realizations of two point processes with proportional intensity functions; meaning they share density function, and therefore have the same spatial structure (in terms of first-order properties). In Chapter 5 an exhaustive simulation study is performed, analysing in various scenarios the quality and well-behaving of the contrast methods proposed in the previous chapter. In order to do so, results of both level and power of those test are presented in different linear networks. Finally, some of the proposed methods have been applied so as to contrast two populations, based on a real-life dataset of trafic accidents in Río de Janeiro (Brasil) from 2019 up to 2022. Moreover, in this work's appendix the code which has been used, with guaranteed reproducibility, can be found. It is worth mentioning that the database has not been included due to condidentiality issues
Description
Traballo Fin de Grao en Matemáticas. Curso 2021-2022
Keywords
Bibliographic citation
Relation
Has part
Has version
Is based on
Is part of
Is referenced by
Is version of
Requires
Sponsors
Rights
Atribución-NoComercial-CompartirIgual 4.0 Internacional



