MODELOS EM ÁRVORE DE DECISÃO PARA ALERTA DA FERRUGEM DO CAFEEIRO
Arquivos
Data
2009
Título da Revista
ISSN da Revista
Título de Volume
Editor
Resumo
O objetivo deste trabalho foi desenvolver modelos em árvore de decisão para alerta da ferrugem do cafeeiro. Dados de oito anos de incidência mensal da doença no campo foram transformados em valores binários, considerando o limite de 5 pontos percentuais na taxa de infecção. Dois modelos foram gerados a partir de dados meteorológicos e do espaçamento entre plantas,, sendo um para lavouras com alta carga pendente de frutos e outro para lavouras com baixa carga pendente. O alerta é indicado quando a taxa de infecção, prevista para o prazo de um mês, atingir ou ultrapassar 5 pontos percentuais. A acurácia do modelo para lavouras com alta carga pendente foi de 81%, calculada por validação cruzada. Este modelo apresentou bons resultados também para outras medidas de avaliação importantes, como sensitividade (80%), especificidade (83%) e confiabilidades positiva (79%) e negativa (84%). O modelo para lavouras com baixa carga pendente não apresentou o mesmo bom desempenho. A acurácia foi estimada em 72% e não houve equilíbrio entre as medidas de avaliação. O modelo de alerta para lavouras com alta carga pendente pode auxiliar na tomada de decisão referente ao controle da ferrugem do cafeeiro no campo. O modelo em árvore de decisão facilita a interpretação e a compreensão de suas regras e assim pode contribuir para que o tomador de decisões tenha maior confiança em adotá-lo como ferramenta de apoio.
The objective of this work was to develop decision tree models for coffee rust warning. Monthly data of disease incidence in the field collected during eight years were transformed into binary values considering the limit of 5 percentage points in the infection rate. Two models were generated from meteorological data and space between plants, one for growing areas with large fruit load and the other for growing areas with small fruit load. The warning is indicated when the infection rate is expected to reach or exceed 5 percentage points in a month. The accuracy by cross validation of the model for growing areas with large fruit load was 81%. This model also showed good results for other important evaluation measures, as sensitivity (80%), specificity (83%), positive reliability (79%) and negative reliability (84%). The model for growing areas with small fruit load did not have the same good performance. It had an estimated accuracy of 79% and did not show the same balance among the evaluation measures. The warning model for growing areas with large fruit load can aid the decision making related to coffee rust control in the field. The decision tree model facilitates the interpretation and understanding of its rules and thus can contribute to increase the decision maker confidence for using that model as a support tool.
The objective of this work was to develop decision tree models for coffee rust warning. Monthly data of disease incidence in the field collected during eight years were transformed into binary values considering the limit of 5 percentage points in the infection rate. Two models were generated from meteorological data and space between plants, one for growing areas with large fruit load and the other for growing areas with small fruit load. The warning is indicated when the infection rate is expected to reach or exceed 5 percentage points in a month. The accuracy by cross validation of the model for growing areas with large fruit load was 81%. This model also showed good results for other important evaluation measures, as sensitivity (80%), specificity (83%), positive reliability (79%) and negative reliability (84%). The model for growing areas with small fruit load did not have the same good performance. It had an estimated accuracy of 79% and did not show the same balance among the evaluation measures. The warning model for growing areas with large fruit load can aid the decision making related to coffee rust control in the field. The decision tree model facilitates the interpretation and understanding of its rules and thus can contribute to increase the decision maker confidence for using that model as a support tool.
Descrição
Trabalho apresentado no Simpósio de Pesquisa dos Cafés do Brasil (6. : 2009 : Vitória, ES). Anais Brasília, D.F: Embrapa - Café, 2011
Palavras-chave
Coffea arabica, Hemileia vastatrix, modelos, predição de doenças de plantas, mineração de dados., Coffea arabica, Hemileia vastatrix, models, plant disease prediction, data mining.
Citação
Meira, Carlos Alberto Alves; Rodrigues, Luiz Henrique Antunes. Modelos em árvore de decisão para alerta da ferrugem do cafeeiro. In: Simpósio de Pesquisa dos cafés do Brasil (6. : 2009 : Vitória, ES). Anais Brasília, D.F: Embrapa - Café, 2011 (1 CD-ROM), 7p.