Table 3.

The classification efficiency values and total accuracy of low-, mid-, and high-level data fusion strategies decision making with PLS-DA and RF models. RFE: Recursive feature elimination, Bo: Boruta, PCs: Principal components, VIP: Variable importance in the projection.

	Model	Calibration Set						Validation Set
		Class1	Class2	Class3	Class4	Class5	Accuracy	Class1	Class2	Class3	Class4	Class5	Accuracy
Low-level	PLS-DA	1.000	1.000	1.000	1.000	1.000	100%	0.926	1.000	0.949	1.000	0.981	95.59%
	RF	0.872	0.885	0.858	0.897	0.932	82.81%	0.886	1.000	0.905	0.972	0.991	92.65%
Low-level (VIP)	PLS-DA	1.000	1.000	1.000	1.000	1.000	100%	0.926	1.000	0.991	1.000	0.991	97.06%
	RF	0.927	0.927	0.849	0.922	0.947	86.72%	0.926	1.000	0.991	1.000	0.991	97.06%
Mid-level (RFE)	PLS-DA	0.946	0.712	0.764	0.864	0.878	75%	0.705	0.794	0.000	0.727	0.798	55.88%
	RF	0.966	0.888	0.868	0.894	0.881	84.38%	0.764	0.861	0.491	0.900	0.854	69.12%
Mid-level (Bo)	PLS-DA	0.961	1.000	1.000	0.995	0.995	98.44%	0.926	1.000	0.991	0.991	1.000	97.06%
	RF	0.947	0.942	0.885	0.922	0.951	89.06%	0.926	1.000	0.949	0.991	0.991	95.59%
Mid-level (PCs)	PLS-DA	0.951	0.961	0.974	0.995	0.995	96.09%	1.000	1.000	0.957	0.991	1.000	98.53%
	RF	0.927	0.951	0.922	0.922	0.981	90.63%	0.886	1.000	0.991	0.981	0.955	94.12%
High-level (RFE)	PLS-DA	0.951	0.981	0.953	1.000	0.990	96.09%	0.870	1.000	0.802	0.991	0.981	89.71%
	RF	0.976	0.976	0.904	0.922	0.995	91.21%	0.926	1.000	0.991	1.000	0.991	97.06%
High-level (Bo)	PLS-DA	0.976	0.981	0.979	1.000	0.990	97.66%	0.926	0.991	0.850	1.000	0.981	92.65%
	RF	0.966	0.976	0.904	0.902	0.951	90.63%	0.917	0.964	0.850	0.972	1.000	91.18%
High-level (PCs)	PLS-DA	0.981	1.000	0.990	0.981	1.000	98.44%	0.964	1.000	1.000	0.991	1.000	98.53%
	RF	0.881	0.971	0.872	0.911	0.961	87.5%	0.966	1.000	1.000	0.991	1.000	100%