Examinando por Autor "Ortiz Rios, Juan Carlos"
Mostrando 1 - 1 de 1
Resultados por página
Opciones de ordenación
- ÍtemSegmentación de suelos de acuerdo con sus características fisicoquímicas a través modelos de aprendizaje automático(Fundación Universitaria Los Libertadores. Sede Bogotá., ) Ortiz Rios, Juan Carlos; González Veloza, José John FredyThe segmentation of the physicochemical quality of the soil makes it possible to establish areas that require similar management or areas with vulnerabilities in which strategies for their conservation and/or recovery should be focused; In this sense, 1139 soil samples were taken from properties in the rural area of the municipalities of Córdoba, Cuaspud, Iles, Ipiales and Potosí in the department of Nariño, on which physicochemical analyzes were carried out (contents of sand, silt and clay, pH, electrical conductivity, organic matter content, nitrogen, exchangeable phosphorus, sulfur, calcium, magnesium, potassium, effective cation exchange capacity, aluminum, iron, manganese, copper, zinc, boron, aluminum saturation, magnesium saturation, saturation potassium, calcium saturation, calcium and magnesium ratio, calcium and potassium ratio, magnesium and potassium ratio, calcium, magnesium and potassium ratio); To establish how these samples could be segmented, a Pearson correlation was initially performed to determine linear relationships between variables, and then a principal component analysis (PCA) was implemented; With this information, several unsupervised learning models were applied to determine the optimal number of clusters in which to segment the information; subsequently, it was decided to carry out a supervised model, Random Forest (RF), taking into account the information from the PCA and clusters, to determine the original variables with greater relative importance in the grouping of information; finally, it was possible to establish the variables and their values that allowed the grouping by applying the Decision Tree (DT) model; In this sense, it was possible to establish that the best way to segment the information of the soil samples is through three clusters, and that the variables that have the greatest weight in the generation of these groups are the contents of Sand and Silt, the Ca/Mg/K ratio and the Ca/K ratio, denoting differences mainly between sandy loam and loam soils.