Identification of patterns for increasing production with decision trees in sugarcane mill data

Sugarcane mills in Brazil collect a vast amount of data relating to production on an annual basis. The analysis of this type of database is complex, especially when factors relating to varieties, climate, detailed management techniques, and edaphic conditions are taken into account. The aim of this paper was to perform a decision tree analysis of a detailed database from a production unit and to evaluate the actionable patterns found in terms of their usefulness for increasing production. The decision tree revealed interpretable patterns relating to sugarcane yield (R2 = 0.617), certain of which were actionable and had been previously studied and reported in the literature. Based on two actionable patterns relating to soil chemistry, intervention which will increase production by almost 2 % were suitable for recommendation. The method was successful in reproducing the knowledge of experts of the factors which influence sugarcane yield, and the decision trees can support the decision-making process in the context of production and the formulation of hypotheses for specific experiments.

Identification of patterns for increasing production with decision trees in sugarcane mill data

Sugarcane mills in Brazil collect a vast amount of data relating to production on an annual basis. The analysis of this type of database is complex, especially when factors relating to varieties, climate, detailed management techniques, and edaphic conditions are taken into account. The aim of this paper was to perform a decision tree analysis of a detailed database from a production unit and to evaluate the actionable patterns found in terms of their usefulness for increasing production. The decision tree revealed interpretable patterns relating to sugarcane yield (R2 = 0.617), certain of which were actionable and had been previously studied and reported in the literature. Based on two actionable patterns relating to soil chemistry, intervention which will increase production by almost 2 % were suitable for recommendation. The method was successful in reproducing the knowledge of experts of the factors which influence sugarcane yield, and the decision trees can support the decision-making process in the context of production and the formulation of hypotheses for specific experiments.(AU)

How accurate are pedotransfer functions for bulk density for Brazilian soils?

The aim of this study was to evaluate the performance of pedotransfer functions (PTFs) available in the literature to estimate soil bulk density (b) in different regions of Brazil, using different metrics. The predictive capacity of 25 PTFs was evaluated using the mean absolute error (MAE), mean error (ME), root mean squared error (RMSE), coefficient of determination (R2) and the regression error characteristic (REC) curve. The models performed differently when comparing observed and estimated b values. In general, the PTFs showed a performance close to the mean value of the bulk density data, considered as the simplest possible estimation of an attribute and used as a parameter to compare the performance of existing models (null model). The models developed by Benites et al. (2007) (BEN-C) and by Manrique and Jones (1991) (M&J-B) presented the best results. The separation of data into two layers according to depth (0-10 cm and 10-30 cm) demonstrated better performance in the 10-30 cm layer. The REC curve allowed for a simple and visual evaluation of the PTFs.(AU)

How accurate are pedotransfer functions for bulk density for Brazilian soils?

The aim of this study was to evaluate the performance of pedotransfer functions (PTFs) available in the literature to estimate soil bulk density (b) in different regions of Brazil, using different metrics. The predictive capacity of 25 PTFs was evaluated using the mean absolute error (MAE), mean error (ME), root mean squared error (RMSE), coefficient of determination (R2) and the regression error characteristic (REC) curve. The models performed differently when comparing observed and estimated b values. In general, the PTFs showed a performance close to the mean value of the bulk density data, considered as the simplest possible estimation of an attribute and used as a parameter to compare the performance of existing models (null model). The models developed by Benites et al. (2007) (BEN-C) and by Manrique and Jones (1991) (M&J-B) presented the best results. The separation of data into two layers according to depth (0-10 cm and 10-30 cm) demonstrated better performance in the 10-30 cm layer. The REC curve allowed for a simple and visual evaluation of the PTFs.