Table 4.

Validated reclassification scores of A. indica S1 and S2 after correction for prior probabilities, as based on the scutum shape (from either landmarks or outline) and on the sensilla shape (outline).

Organ and method used	No used (n1, n2)	Clustering	Validated supervised reclassification
		K-Means (K = 2)	Discriminant analysis	Random Forest
Sensilla outline	60, 64	96 % (optK = 2)	99 % (17PCs)	97 % (17PCs)
Scutum outline	60, 64	53 % (optK = 7)	43 % (17PCs)	40 % (17PCs)
Scutum landmarks	62, 62	64 % (optK = 4)	42 % (10PCs)	42 % (10PCs)

Legend of Table 4. Percentages of correctly assigned individuals to either S1 (club-shaped sensilla) or S2 (rounded sensilla) morphotypes (Fig. 1), according to unsupervised and supervised reclassification and after adjustment for prior probabilities (see Materials and Methods). The K-means analysis was performed with K = 2, using the total number of shape variables; opt K, the optimum number of groups (see Materials and Methods for more details). The random forest algorithm used 400 bootstraps. The set of first principal components (PCs) used as input are indicated between brackets: they were PCs of shape variables (excluding size) and represented together at least 95 % of the total shape variation.