Table 3.

Classification Performance by Retinal Oversampling Before and After Applying Feature Selection

	Before Feature Selection			After Feature Selection
Method	ACC	SEN	SPE	ACC	SEN	SPE
LDA	90.9 ± 0.0%	91.1 ± 3.1%	93.4 ± 1.2%	89.1 ± 0.3%	84.5 ± 3.4%	94.4 ± 1.6%
SVM	92.5 ± 0.3%	95.6 ± 0.7%	91.4 ± 0.7%	92.4 ± 0.3%	94.0 ± 1.4%	92.0 ± 2.1%
RF	93.7 ± 0.2%	95.5 ± 1.3%	92.1 ± 2.1%	92.8 ± 0.5%	94.9 ± 0.7%	91.4 ± 2.4%

The average accuracy (ACC), sensitivity (SEN), and specificity (SPE) ± standard deviation before feature selection (with all polarimetric properties included) and after feature selection (including a small subset with high variable importance shown in Figure 5) for each of the 3 classifiers: linear discriminant analysis (LDA), support vector machine (SVM) and random forest (RF) for the retinal oversampling method.