SpringerPlus

Table 8 Contrasting performance for randomly reordered data

From: Vertex collocation profiles: theory, computation, and results

(a) AUROC
Network	VCP ^3,1,0	VCP ^3,2,0	VCP ^3,2,0 Reordered	VCP ^4,1,0	VCP ^4,2,0	VCP ^4,2,0 Reordered
condmat	0.637	0.670	0.632	0.582	0.646	0.589
dblp-collab	0.640	0.749	0.682	0.695	0.789	0.707
hepth-collab	0.598	0.738	0.599	0.622	0.732	0.596
huddle	0.881	0.929	0.886	0.888	0.926	0.888
(b) AUPR
Network	VCP ^3,1,0	VCP ^3,2,0	VCP ^3,2,0 Reordered	VCP ^4,1,0	VCP ^4,2,0	VCP ^4,2,0 Reordered
condmat	0.011917	0.010643	0.011001	0.008588	0.010103	0.009829
dblp-collab	0.007152	0.009161	0.008131	0.009410	0.012389	0.008881
hepth-collab	0.007197	0.010280	0.006494	0.007157	0.009375	0.006622
huddle	0.039394	0.065830	0.044412	0.046803	0.071811	0.046162

Columns showing performance with reordered data represent the mean performance over 10 random orderings of training data. Bold values indicate temporally ordered data outperforming randomly ordered data with statistical significance at 99.9% confidence.

Back to article page