CC brings greatest improvements to the predictions of burst count (BC) and burst length (BL) in networks with binomial in-degree distribution.

<p><b>Left:</b> The y-axis shows the relative improvements with respect to null prediction. For each simulation setting, the prediction error for null predictor and the predictor with a considered graph property are calculated, using and . The relative improvements are averaged over all 12 simulation settings with binomial in-degree distribution. Plotted is the improvement (mean and std) for repetitions. The improvement obtained by using CC (*) in the prediction is significantly greater than that obtained by any other single graph measure. <b>Right:</b> The y-axis shows relative improvements with respect to prediction by other graph properties. As an example, the first bar shows the relative improvement , averaged over graph properties NB,OD,MEig,Mot5,Mot12. The improvements are further averaged over all 12 simulation settings, and the mean + std of repetitions are shown. The procedure is similar for the other bars. The improvement obtained by using CC (*) in the coprediction is significantly greater than that obtained by any other graph measure (U-test, ).