Please use this identifier to cite or link to this item:
Title: Hipsometria: seleção de variáveis e mineração de dados por métodos de inteligência computacional
Other Titles: Hypsometry: feauture selection and data mining by computer intelligence methods
Authors: Gomide, Lucas Rezende
Barbosa, Bruno Henrique Groenner
Gomide, Lucas Rezende
Silva, Carolina Souza Jarochinski e
Scolforo, Henrique Ferraço
Keywords: Seleção de recursos
Algoritmo genético
Florestas nativas
Seleção de variáveis
Feauture selection
Genetic algorithm
Native forests
Issue Date: 11-Mar-2020
Publisher: Universidade Federal de Lavras
Citation: MIRANDA, E. N. Hipsometria: seleção de variáveis e mineração de dados por métodos de inteligência computacional. 2020. 87 p. Dissertação (Mestrado em Engenharia Florestal)–Universidade Federal de Lavras, Lavras, 2020.
Abstract: The assertive use of dendrometric variables has a direct impact on forest planning, so more accurate results are needed. Height stands out among biometric variables as it is an important attribute commonly used for volume calculation methods and for measuring height and volume increment of trees, etc. Thus, new technologies and techniques have been implemented in recent years to assist in their calculation. In the context of height estimation, traditional statistical models that have a good response can be improved with data mining techniques. In this dissertation, the principle of data mining was applied to feature selection in both chapters to estimate the individual height of the trees of the Rio Grande basin - MG. In the first chapter, the objective was to select variables within traditional literature models, where possible variable combinations were applied as input to nonlinear models, using a dual genetic algorithm, the first selects and assembles the variable combinations, the second parameterize and adjust the constructed model. The generated models presented a small gain in the estimates, and the proposed methodology proved to be efficient in the search for good results, but with difficulties to find good results in problems with many inputs. The proposal proved robust and can be applied to other problems. The second chapter sought to compare traditional methods of predicting height with machine learning methods in their pure and hybrid form. The Random Forest (RF) model with variable reduction proved to be robust, capable of improving the response and reducing the number of entries in the RF model, presenting better results to the others. Techniques that involve the use of computational intelligence are effective in the search for good results, with superior answers than traditional ones, capable of selecting good variables and estimating good height values.
Appears in Collections:Engenharia Florestal - Mestrado (Dissertações)

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.