Services on Demand
Journal
Article
Indicators
- Cited by SciELO
Related links
- Similars in SciELO
Share
Agriscientia
On-line version ISSN 1668-298X
Abstract
SUAREZ, F; BRUNO, C.; GIMENEZ PECCI, M. P and BALZARINI, M.. Comparación de procedimientos de selección de variables para la modelación de la relación clima-patógenos en cultivos. Agriscientia [online]. 2023, vol.40, n.2, pp.4-4. ISSN 1668-298X. http://dx.doi.org/10.31047/1668.298x.v40.n2.40871.
Nowadays it is possible to easily access large volumes of georeferenced climatic data. These data can be used to model the relationship between climatic conditions and disease from múltiple meteorológica! variables, usually correlated and redundant. The selection of variables allows the identificaron of a subset of relevant regressors to build predictive models. Stepwise, Boruta, and LASSO are variable selection procedures of different nature, so their relative performance has been scarcely explored. The objective of this work was the comparison of these methods simultaneously applied in the construction of regression models to predict disease risk from climatic data. Three georeferenced databases were used with presence/absence values of different pathogens in maize crops in Argentina. For each scenario, climatic variables from the period prior to sowing until harvest were obtained. The three variable selection methods obtained models with accuracy close to 70 %. However, LASSO produced the best predictive model, selecting an intermediate number of variables with respect to Stepwise (lower number) and Boruta (higher number). The results could be extended to other pathosystems and inspire the construction of alarm systems based on climatic variables.
Keywords : LASSO; Stepwise; Boruta; logistic regression.