Figure 4

The feature space for the distance-based approach. To compare the two approaches, we visualize the feature space of the two top-ranked features by ANOVA. (a) represents feature space of the descriptor-based approach where $F_{i=1,2}^{desc}$. (b) represents feature space of the non distance-based approach where $F_{i=1,2.}^{deg}$. ^deg. To gain the features vectors in (b), we calculate the Kullback-Leibler divergence (KLD) based on the degree distribution between each network. This leads to a feature vector for each network containing the KLD for each of the other networks.