Publication: Aplicación de técnicas de minería de datos geo-referenciados en los circuitos de comercialización alternativa de productos agrícolas en Ecuador
Loading...
Identifiers
Publication date
2019-09-19
Defense date
2019-07-16
Authors
Tutors
Journal Title
Journal ISSN
Volume Title
Publisher
Impact
Export
Abstract
A nivel mundial se utilizan sistemas de información para realizar el seguimiento y optimización de la producción agrícola.
En el Ecuador el ministerio de Agricultura y Ganaderia (MAG), tiene un pro-grama orientado a fortalecer la asociación de productores agrícolas familiares que comercializan sus productos de manera directa con el consumidor, en un denominado circuito alternativo de comercialización (CIALCO).
A la información recolectada por el MAG, de ferias tipo Cialco, ubicadas en las provincias de Tungurahua y Chimborazo, se aplican técnicas de minería de datos descriptivas y predictivas, para descubrir patrones de comportamiento que permitan optimizar la utilización del suelo y mejorar los ingreso en la comercialización de productos agrícolas de este sector.
En la parte descriptiva, basados en la inducción de reglas de asociación, generadas utilizando los algoritmos A priori y FP-growth con parámetros mínimos de soporte y confianza, se genera un conjunto que se compone de todos los elementos resultado de obtener las mejores reglas.
El conjunto asociativo resultante se integra por los productos cebolla blanca, tomate de árbol, zanahoria, brócoli y tomate riñón.
En la parte predictiva se busca realizar una estimación pronóstica utilizando dos dimensiones: tiempo y ubicación geográfica.
Con un solo predictor, se genera una serie de tiempo utilizando el algoritmo SMOReg, para realizar una extrapolación pronostica con la que se encuentra valores de comercialización de productos agrícolas fuera del periodo de registro de información.
Adicionando coordenadas geográficas a la información inicial se ubican espacialmente las ferias en la región de estudio, compuesto por las provincias de Tungurahua y Chimborazo, para utilizar la dimensión espacial y en base a procesos de kriging realizar interpolación pronóstica para estimar va-lores de comercialización en lugares donde no se tiene información.
Una vez desarrollados estos tres procesos de minería de datos se propone una metodología qué, utilizando el conjunto asociativo como predictor, vuelve a calcular la estimación pronostica para la dimensión tiempo y la dimensión espacio.
La comparación de resultados con un solo predictor frente a los resultados de estimación pronóstica utilizando el conjunto asociativo como predictor indican que los porcentajes de error en la estimación pronostica multivariable disminuyen de manera considerable.
Para validar los resultados obtenidos de mejora de estimación pronostica, se crean dos modelos de datos utilizando variables externas al proceso de comercialización población y piso climático.
En los resultados finales, se aprecia que las dos variables de forma independiente muy poco aportan en la disminución del error de estimación, mientras que si se las hace interactuar con el conjunto asociativo se vuelve a encontrar una disminución en el error de estimación pronóstica obtenido.
At the world level, information systems are used to monitor and optimize agricultural production. In Ecuador, the Ministry of Agriculture and Livestock has a program aimed at strengthening the association of family agricultural producers, who market their products directly with the consumer, in a socalled alternative marketing circuit (CIALCO). To the information collected from Cialcos-type fairs, located in the provinces of Tungurahua and Chimborazo, descriptive and predictive data mining techniques are applied. To discover patterns of behavior that allow to optimize the use of the soil and improve the income in the commercialization of agricultural products. In the descriptive part, based on the induction of association rules, generated using the Apriori and FP-growth algorithms with minimum support and Confidence parameters, a set is generated that consists of all the elements resulting from obtaining the best rules. The resulting associative set is integrated by the products white onion, tree tomato, carrot, broccoli and tomato kidney. The resulting associative set is integrated by the products: white onion, tree tomato, carrot, broccoli and tomato kidney. In the predictive part, a prognostic estimation is sought using two dimensions: time and geographic location. With a single predictor, a series of time is generated using the SMOReg algorithm, to perform a forecast extrapolation with which commercialization values of agricultural products are found outside the period of information registration. By adding geographical coordinates to the initial information, the fairs are located spatially in the study region, composed of the provinces of Tungurahua and Chimborazo, to use the spatial dimension and based on kriging processes to perform prognostic interpolation to estimate marketing values in places where you do not have information. Once these three processes of data mining have been developed, it is proposed to establish a methodology that, using the associative set as a predic-tor, recalculates the forecast forecast for the time dimension and the space dimension. The comparison of results with a single predictor versus the results of prognostic estimation using the associative set as a predictor they indicate that the percentages of error in the multivariable forecast estimate decrease considerably. In order to validate the results obtained from improvement of forecast estimation, two data models are created using variables external to the population and climatic floor marketing process. In the final results, it can be seen that the two variables independently contribute very little in the reduction of the estimation error, whereas if they are made to interact with the associative set, they will find a decrease in the error obtained.
At the world level, information systems are used to monitor and optimize agricultural production. In Ecuador, the Ministry of Agriculture and Livestock has a program aimed at strengthening the association of family agricultural producers, who market their products directly with the consumer, in a socalled alternative marketing circuit (CIALCO). To the information collected from Cialcos-type fairs, located in the provinces of Tungurahua and Chimborazo, descriptive and predictive data mining techniques are applied. To discover patterns of behavior that allow to optimize the use of the soil and improve the income in the commercialization of agricultural products. In the descriptive part, based on the induction of association rules, generated using the Apriori and FP-growth algorithms with minimum support and Confidence parameters, a set is generated that consists of all the elements resulting from obtaining the best rules. The resulting associative set is integrated by the products white onion, tree tomato, carrot, broccoli and tomato kidney. The resulting associative set is integrated by the products: white onion, tree tomato, carrot, broccoli and tomato kidney. In the predictive part, a prognostic estimation is sought using two dimensions: time and geographic location. With a single predictor, a series of time is generated using the SMOReg algorithm, to perform a forecast extrapolation with which commercialization values of agricultural products are found outside the period of information registration. By adding geographical coordinates to the initial information, the fairs are located spatially in the study region, composed of the provinces of Tungurahua and Chimborazo, to use the spatial dimension and based on kriging processes to perform prognostic interpolation to estimate marketing values in places where you do not have information. Once these three processes of data mining have been developed, it is proposed to establish a methodology that, using the associative set as a predic-tor, recalculates the forecast forecast for the time dimension and the space dimension. The comparison of results with a single predictor versus the results of prognostic estimation using the associative set as a predictor they indicate that the percentages of error in the multivariable forecast estimate decrease considerably. In order to validate the results obtained from improvement of forecast estimation, two data models are created using variables external to the population and climatic floor marketing process. In the final results, it can be seen that the two variables independently contribute very little in the reduction of the estimation error, whereas if they are made to interact with the associative set, they will find a decrease in the error obtained.
Description
Keywords
Minería de datos, Series temporales, Pronóstico (Estadística), Extrapolación, Regresión lineal, Variable aleatoria, Variograma, Kriging, Cokriging, Fusión de datos, Reglas de aociación, SDG, ODS