Table 3. Classification Model Characteristics and Correctly Predicted Test Set Matrix Samples^a,b.

			OPLS-DA						RF
classification model	attribution capacity	class	comp.a	R2X (cum)	R2Y (cum)	Q2(cum)	pred. soil	pred. textile	OOB error (%)	pred. soil	pred. textile
M1matrix	ethylene or TDG routes	R(10, 11)	1 + 2 + 0	0.45	0.99	0.97	-	-	0	-	-
		R(1–9)
M2matrix	chlorination methods	R(1, 4, 7)	2 + 3 + 0	0.46	0.97	0.8	5/6	4/4b	0	5/6	4/4b
		R(2, 5, 8)					5/6	5/5b		5/6	5/5b
		R(3, 6, 9)					2/6	2/6		2/6	2/6
M3amatrix	TDG synthesis methods of R(1, 4, 7) samples	R1	2 + 2 + 0	0.75	0.93	0.66	2/2	0/1	2.9	2/2	1/1
		R4					0/2	0/1		0/2	0/1
		R7					2/2	2/2		2/2	2/2
M3bmatrix	TDG synthesis methods of R(2, 5, 8) samples	R2	2 + 3 + 0	0.80	0.97	0.91	2/2	1/1	2.8	0/2	0/1
		R5					2/2	2/2		2/2	2/2
		R8					2/2	2/2		2/2	2/2
M3cmatrix	TDG synthesis methods of R(3, 6, 9) samples	R3	2 + 0 + 0	0.73	0.64	0.26	0/2	0/2	0	0/2	0/2
		R6					2/2	2/2		2/2	2/2
		R9					2/2	0/2		2/2	2/2

^aComp. shows the number of components (x/y joint predictive variation + variation in x orthogonal to y + variation in y orthogonal to x) included in each model.

^bIn the initial PCA model, two outliers were detected among the training set samples and three in the test set and thus excluded from further analysis.