Classification of soil respiration in areas of sugarcane renewal using decision treecultural

Abstract

The use of data mining is a promising alternative to predict soil respiration from correlated variables. Our objective was to build a model using variable selection and decision tree induction to predict different levels of soil respiration, taking into account physical, chemical and microbiological variables of soil as well as precipitation in renewal of sugarcane areas. The original dataset was composed of 19 variables (18 independent variables and one dependent (or response) variable). The variable-target refers to soil respiration as the target classification. Due to a large number of variables, a procedure for variable selection was conducted to remove those with low correlation with the variable-target. For that purpose, four approaches of variable selection were evaluated: no variable selection, correlation-based feature selection (CFS), chisquare method (chi(2)) and Wrapper. To classify soil respiration, we used the decision tree induction technique available in the Weka software package. Our results showed that data mining techniques allow the development of a model for soil respiration classification with accuracy of 81 %, resulting in a knowledge base composed of 27 rules for prediction of soil respiration. In particular, the wrapper method for variable selection identified a subset of only five variables out of 18 available in the original dataset, and they had the following order of influence in determining soil respiration: soil temperature > precipitation > macroporosity > soil moisture > potential acidity.