Tracing Geographical Origins of Teas Based on FT-NIR Spectroscopy: Introduction of Model Updating and Imbalanced Data Handling Approaches

Table 4

Comparison of prediction results for 2015 dataset based on OC-SVM and IF combined SVM classifiers at different updating ratios.

Updating rate	One-class SVM (OC-SVM)			Isolation forest (IF)
	Recall 1	Recall 2	MAR^a	Recall 1	Recall 2	MAR
10%	0.8275	0.7324	0.7799 G	0.8414	0.7257	0.7835 G
20%	0.8482	0.8386	0.8434 E	0.8617	0.7983	0.8300 F
30%	0.8612	0.8749	0.8680 D	0.8717	0.8312	0.8515 E
40%	0.8739	0.9077	0.8908 B	0.8767	0.8599	0.8683 D
50%	0.8824	0.9238	0.9031 A	0.8802	0.8727	0.8765 C,D
60%	0.8867	0.9379	0.9123 A	0.8860	0.8770	0.8815 B,C

^aMAR: macro average recall. Means with the same letter(s) are not significantly different at 0.01 level.