CLASSIFICATION OF SUGARCANE YIELDS ACCORDING TO SOIL FERTILITY PROPERTIES USING SUPERVISED MACHINE LEARNING METHODS

Yepes, Jhonnatan; Oré, Gian; Alcântara, Marlon S.; Hernandez-Figueroa, Hugo E.; Teruel, Bárbara

doi:10.1590/1809-4430-Eng.Agric.v42n5e20210239/2022

Acessibilidade / Reportar erro

Brasil

Español English

sumário « anterior atual seguinte »

Sumário

Scientific Paper • Eng. agríc. 42 (5) • 2022 • https://doi.org/10.1590/1809-4430-Eng.Agric.v42n5e20210239/2022 copy

CLASSIFICATION OF SUGARCANE YIELDS ACCORDING TO SOIL FERTILITY PROPERTIES USING SUPERVISED MACHINE LEARNING METHODS

Authorship SCIMAGO INSTITUTIONS RANKINGS

ABSTRACT

Action planning and decision-making in the sugarcane management chain depend on yield estimates, which, in turn, vary with the soil. This study aimed to describe an applicable method of classifying sugarcane productivity into three categories, based on soil properties (medium, low, and high), determining which is most associated with biomass production. To this end, we applied the machine learning methods Naïve Bayes, Decision Trees, and Random Forest, as they proved to be useful tools for faster and more accurate results. Our results indicate that Random Forest is the most suitable for classifying all yield categories, and Naïve Bayes had good results for classification into “medium” and “low” and potential for solving multiclass problems in agriculture. Organic matter was the property most closely related to sugarcane biomass yield by the Random Forest and Decision Trees algorithms. The methods described can be used to obtain subsidies for sugarcane chain management, contributing to more sustainable decisions.

random forest; naïve bayes; decision trees; multi-class classification; classification techniques

Algorithm	Parameter
Naïve Bayes	Batch size= 100
	Kernel estimator: Yes
	Normalized values: Yes
Decision Trees	Criterion: Entropy
	Minimum number of samples required to split an internal node= 2
	Minimum number of samples required to be at a leaf node= 5
	Maximum leaf nodes= 5
	Class weight: Balanced
	Normalized values: Yes
Random Forest	Criterion: Entropy
	Number of trees in the forest= 50
	Minimum number of samples required to split an internal node= 2
	Minimum number of samples required to be at a leaf node= 5
	Maximum leaf nodes= 5
	Class weight: Balanced
	Normalized values: Yes

	OM	pH	P	K	Ca	Mg	B	Cu	Fe	Mn	Zn
Mean	25.7	4.53	11.3	12.9	90.8	6.1	0.30	15.1	18.6	22.6	4.1
Standard deviation	1.7	0.18	2.7	2.5	30.4	1.4	0.006	3.5	3.3	0.4	2.5
Standard error	0.14	0.02	0.23	0.21	2.53	0.12	0.0005	0.29	0.28	0.03	0.21
Lower value	22.3	4.2	8.2	9.0	44.0	3.6	0.29	9.0	13.2	21.3	0.8
25%	24.4	4.3	8.7	10.8	62.0	4.8	0.296	11.7	16.1	22.5	1.7
50%	25.8	4.5	10.7	12.9	86.4	6.0	0.3	15.0	18.9	22.7	3.2
75%	27.5	4.7	13.9	15.2	119.0	7.3	0.30	18.1	20.6	22.8	6.3
Maximum value	28.2	4.8	16.5	16.8	144.0	8.6	0.31	21.0	26.5	23.3	9.8

Ca²⁺ (mg dm^-3)
Dec/19	Jan/20	Feb/20	Mar/20	Apr/20	May/20
98.5	90.0	85.0	86.5	88.5	81.5

Naïve Bayes
Class	True positive rate	False positive rate	Accuracy	Recall	AUC
High	0.36	0.18	0.40	0.36	0.70
Medium	0.40	0.22	0.62	0.40	0.70
Low	0.83	0.32	0.50	0.83	0.84
Weighted average	0.51	0.24	0.53	0.51	0.74

Decisions Tree
Class	True positive rate	False positive rate	Accuracy	Recall	AUC
High	1.00	0.78	0.30	1.00	0.47
Medium	0.00	0.00	--	--	--
Low	0.50	0.03	0.86	0.50	0.63
Weighted average	0.40	0.21	--	0.40	--

Random Forest
Class	True positive rate	False positive rate	Accuracy	Recall	AUC
High	0.64	0.28	0.44	0.64	0.70
Medium	0.35	0.26	0.54	0.35	0.52
Low	0.75	0.16	0.64	0.75	0.79
Weighted average	0.58	0.24	0.54	0.54	0.64

Metrics	Algorithm
Metrics	Naïve Bayes	Decision Trees	Random Forest
Correctly classified instances (%)	51.16	39.53	53.48
Incorrectly classified instances (%)	48.84	60.46	46.51
Kappa coefficient	0.27	0.18	0.30
MAE	0.34	0.40	0.38
RMSE	0.47	0.44	0.47
RAE	0.77	0.89	0.85

Associação Brasileira de Engenharia Agrícola SBEA - Associação Brasileira de Engenharia Agrícola, Departamento de Engenharia e Ciências Exatas FCAV/UNESP, Prof. Paulo Donato Castellane, km 5, 14884.900 | Jaboticabal - SP, Tel./Fax: +55 16 3209 7619 - Jaboticabal - SP - Brazil
E-mail: revistasbea@sbea.org.br

Acompanhe os números deste periódico no seu leitor de RSS

[1] 1* Corresponding author. School of Agricultural Engineering, University of Campinas-UNICAMP/ Campinas - SP, Brasil. E-mail: jayepesg@unal.edu.co