Thesis APLICACIÓN DE TÉCNICAS DE MACHINE LEARNING PARA PREDECIR EL TAMAÑO DE INCENDIOS FORESTALES
Loading...
Date
2017
Authors
CAMPOS SANTELICES, MATÍAS FELIPE
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
El control y contención de incendios forestales se ha convertido en una temática cada vezmás importante alrededor del mundo, no tan solo por el daño que éstos causan al hábitat delser humano, sino que también por el dinero invertido en su combate. Es por esto, que contarcon sistemas que apoyen la toma de decisiones para afrontar estos eventos naturales es fundamentalpara asignar recursos de forma eficiente. El presente trabajo propone un modelopredictivo del tamaño de los incendios forestales utilizando técnicas de machine learning, enel cual mediante el uso del proceso de minería de datos CRISP-DM, se proponen dos acercamientospara resolver dicha tarea: primeramente prediciendo esta área de forma cuantitativa,aplicando un modelo de regresión; y por otra parte, cuantitativamente prediciendo su tamañoa través de etiquetas. En particular se evaluará el desempeño de las maquinas de soportevectorial, evaluando las ventajas y desventajas de ambos acercamientos y comparando losresultados obtenidos con estudios similares.
Controlling and containing forest fires has become an important topic of discussion aroundthe world due to not only the great damaged they cause to human habitat but also the amountsof financial resources invested in their extinction. This is where lies the importance of havingsystems that support decision making on how to efficiently allocate resources to fightthem. This present paper proposes a predictive modeling for the magnitude of forest firesusing machine learning techniques. Through the use of CRISP-DM data mining process twoproposal are presented. First, a quantitative prediction by applying a regression model, andsecond, quantitatively predicting its size through tags. In particular, the performance of SupportVector Machines will be assessed, evaluating the advantages and disadvantages of bothapproaches and comparing the results obtained with similar studies.
Controlling and containing forest fires has become an important topic of discussion aroundthe world due to not only the great damaged they cause to human habitat but also the amountsof financial resources invested in their extinction. This is where lies the importance of havingsystems that support decision making on how to efficiently allocate resources to fightthem. This present paper proposes a predictive modeling for the magnitude of forest firesusing machine learning techniques. Through the use of CRISP-DM data mining process twoproposal are presented. First, a quantitative prediction by applying a regression model, andsecond, quantitatively predicting its size through tags. In particular, the performance of SupportVector Machines will be assessed, evaluating the advantages and disadvantages of bothapproaches and comparing the results obtained with similar studies.
Description
Catalogado desde la version PDF de la tesis.
Keywords
CLUSTERING , CRISP-DM , INCENDIOS FORESTALES , MACHINE LEARNING , MAQUINAS DE SOPORTE VECTORIAL
