Fig. 4.

Nutritional quality biplot and RF model variable importance (a) PCA biplot of the three nutritional quality classes based on phenolic and micronutrient contents wherein PC1 explains 22.6% of the variation while PC2 explains 14.4% of the variation. Phenotypic distribution box plots of the pigmented rice collections based on (b) VideometerLAB parameters (c) phenolics content and (d) mineral content arranged according to cluster (1,2 and 3). Resulting variable importance of the random forest models generated using (f) multi-spectral imaging parameters, (e) the combinations of multi-spectral imaging, phenolic content and mineral content data (g) the top 5 important variable and multi-spectral imaging data, and (h) individual variable contributions per class. Abbreviated variables: L- whiteness, A- redness, B- yellowness, TPC – total phenolic content, TFC – total flavonoid content, TAC- total anthocyanin content, PC- phenolic content, FC- flavonoid content, AC- anthocyanin content, Mn - manganese content, Mo- molybdenum content, Na – sodium content, P – phosphorus content, S- sulphur content, Zn- zinc content, Ca- calcium content, Cu- copper content, Fe- iron content, K- potassium content, Mg- magnesium content.