Fig 8. Feature selection by the embedded method after Feature Concatenation in the VMR datasets.

(A–B) Top 30 features from the light-on dataset matricized by Concat3Dist and Concat9Meas. (C–D) Top 30 features from the light-off dataset matricized by Concat3Dist and Concat9Meas. These features (Y-axis) were named as in Fig 6 and ranked by the Random Forest mechanism using the mean decrease of the Gini index (MeanDecreaseGini; X-axis) in descending order as in Fig 7. The features were color-coded based on the color scheme of features shown in Fig 2A. The selection cutoff thresholds (vertical dash lines) are 0.6, 0.65, 1, and 0.6 for (A) to (D), respectively.