Evaluating soil nutrients of Dacrydium pectinatum in China using machine learning techniques

Forest Ecosystems

Table 5 Accuracy of the sample training set and testing set of the six models

Soil nutrient grade		KNNSVM (%)	LSVM (%)	LMSVM (%)	SVM-k-NN (%)	FSVM (%)	PSVM (%)
Training set ratio	I	96.9	96.5	80.4	92.8	81.7	88.2
	II	97.5	95.8	90.7	93.4	80.8	89.7
	III	96.1	92.9	93.2	92.8	98.3	94.8
	IV	98.5	85.8	98.9	98.0	98.7	92.4
	V	96.0	88.3	80.1	93.3	97.6	95.2
	VI	96.6	91.1	98.8	91.7	96.3	93.8
	Average	96.6	91.7	90.4	93.7	92.9	92.4
Testing set ratio	I	97.0	83.6	87.5	81.1	91.1	92.4
	II	90.0	87.4	90.7	86.5	92.7	90.7
	III	95.5	91.3	88.6	98.2	91.7	93.3
	IV	86.8	82.9	80.4	97.7	87.5	89.1
	V	94.4	87.3	81.3	90.2	88.6	90.4
	VI	98.1	88.5	84.1	97.7	89.7	80.6
	Average	93.6	86.8	85.4	91.9	89.9	88.9

Note: KNNSVM is the k-nearest neighbors local support vector machine; LSVM is the localized support vector machine; LMSVM is the local mixture-based support vector machine; SVM-KNN is the k-NN and SVM integrated algorithm; FSVM is the fast local kernel support vector machine; PSVM is the proximal support vector machine. I, II, III, IV, V, and VI indicate that the soil nutrient quality grade is “extremely high”, “medium high”, “low”, “poor”, and “extremely poor”, respectively