Aplicação de técnicas de machine learning para classificação da aptidão dos solos para o regadio
Type
bookPart
Publisher
Identifier
TORRES, Pedro [et al.] (2022) - Aplicação de técnicas de machine learning para classificação da aptidão dos solos para o regadio. In: Agrárias : pesquisa e inovação nas ciências que alimentam o mundo. Curitiba : ARTEMIS. Vol. VIII, p.209-224. ISBN 978-65-87396-68-2. DOI 10.37572/EdArt_260822682
978-65-87396-68-2
10.37572/EdArt_260822682
Title
Aplicação de técnicas de machine learning para classificação da aptidão dos solos para o regadio
Application of machine learning technics for evaluation of the soils capability to irrigation
Application of machine learning technics for evaluation of the soils capability to irrigation
Subject
Aptidão solos regadio
Machine Learning
Scikit-Learn
Aprendizagem Supervisionada.
Soils capability irrigation
Machine Learning
Scikit-Learn
Supervised Learning
Machine Learning
Scikit-Learn
Aprendizagem Supervisionada.
Soils capability irrigation
Machine Learning
Scikit-Learn
Supervised Learning
Date
2022-10-11T10:49:15Z
2022-10-11T10:49:15Z
2022
2022-10-11T10:49:15Z
2022
Description
Este trabalho consiste no desenvolvimento e validação de modelos de Machine Learning para a otimização de um sistema de rega de precisão utilizando algoritmos de classificação. A finalidade é atribuir a cada solo,
localizado a sul do concelho do Fundão, Portugal, uma classe de aptidão para o regadio, classes essas que identificam as zonas regáveis, não regáveis bem como as que precisam de intervenção para serem regadas. Os dados dos casos de estudo foram anteriormente recolhidos por uma aluna de Mestrado da Escola
Superior Agrária do IPCB (Portugal), onde incluíam vários condicionalismos (características dos solos que podem condicionar a aptidão para o regadio). A análise exploratória dos dados permitiu utilizar apenas os valores dos resultados relativamente às características dos solos que podem condicionar a aptidão para
o regadio rejeitando assim todo o cálculo efetuado para a obtenção dos mesmos. Desta forma os dados do caso de estudo foram enriquecidos com esta informação para a aplicação nos algoritmos de Machine Learning. Em geral, o facto de retirar estas características que não revelavam impacto no estudo ajudaram a melhorar os modelos de classificação bem como a sua precisão. Diferentes algoritmos de Machine Learning foram desenvolvidos, testados e validados, tais como, Support Vetor Machine, kNN, Árvore de Decisão, Naive
Bayes e Regressão Logística, para otimizar um sistema de rega de precisão de modo a atribuir uma a classe de aptidão de rega a novos solos introduzidos. A comparação dos modelos demonstrou que o método Naive Bayes é o que apresenta uma melhor precisão na altura de gerar uma classe de previsão.
This work consists of the development and validation of Machine Learning models for the optimization of a precision irrigation system using classification algorithms. The purpose is to assign to each soil, located in the south of the municipality of Fundão, Portugal, an class of capability to irrigation, classes that identify the irrigable and non-irrigated areas as well as those that need intervention to be irrigated. Data from the case studies were previously collected by a Master's student at the Escola Superior Agrária – IPCB (Portugal), which included several constraints (characteristics of soils that may affect the suitability for irrigation). The exploratory analysis of the data allowed us to use only the values of the results regarding the characteristics of the soils that may affect the suitability for irrigation, thus rejecting all the calculation made to obtain them. In this way, the case study data were enriched with this information for application in Machine Learning algorithms. In general, removing these features that had no impact on the study helped to improve the classification models as well as their accuracy. Different Machine Learning algorithms were developed, tested, and validated, such as Support Vector Machine, kNN, Decision Tree, Naive Bayes and Logistic Regression, to optimize a precision irrigation system in order to assign an irrigation suitability class. to new introduced soils. The comparison of the models showed that the Naive Bayes method is the one that presents the best precision when generating a prediction class.
info:eu-repo/semantics/publishedVersion
This work consists of the development and validation of Machine Learning models for the optimization of a precision irrigation system using classification algorithms. The purpose is to assign to each soil, located in the south of the municipality of Fundão, Portugal, an class of capability to irrigation, classes that identify the irrigable and non-irrigated areas as well as those that need intervention to be irrigated. Data from the case studies were previously collected by a Master's student at the Escola Superior Agrária – IPCB (Portugal), which included several constraints (characteristics of soils that may affect the suitability for irrigation). The exploratory analysis of the data allowed us to use only the values of the results regarding the characteristics of the soils that may affect the suitability for irrigation, thus rejecting all the calculation made to obtain them. In this way, the case study data were enriched with this information for application in Machine Learning algorithms. In general, removing these features that had no impact on the study helped to improve the classification models as well as their accuracy. Different Machine Learning algorithms were developed, tested, and validated, such as Support Vector Machine, kNN, Decision Tree, Naive Bayes and Logistic Regression, to optimize a precision irrigation system in order to assign an irrigation suitability class. to new introduced soils. The comparison of the models showed that the Naive Bayes method is the one that presents the best precision when generating a prediction class.
info:eu-repo/semantics/publishedVersion
Access restrictions
openAccess
Language
por
Comments
