Fingerprinting Bitcoin entities using money flow representation learning

Applied Network Science

Table 2 Entity classification accuracy, F1-measure, and AUROC of SVM models for the top 5% frequent cluster embeddings with different walks and patterns

	Mining Pools			BitcoinHeist			WalletExplorer
	Accuracy	F1	AUROC	Accuracy	F1	AUROC	Accuracy	F1	AUROC
Baseline (Freq. 1% Clusters)
Graph2Vec	0.2693	0.2168	0.8011	0.4124	0.3364	0.8931	0.3779	0.3054	0.8968
SVD	0.6253	0.5576	0.9556	0.3191	0.2706	0.8163	0.3536	0.2785	0.8992
Sequential Pattern (Freq. 5% Clusters)
Shortest Path Walk	0.8800	0.8506	0.9894	0.6346	0.5909	0.9399	0.4903	0.4171	0.9293
Random Walk	0.9493	0.9372	0.9967	0.7023	0.6652	0.9541	0.6282	0.5632	0.9583
Biased Random Walk	0.6373	0.5769	0.9501	0.5945	0.5604	0.9283	0.4424	0.3671	0.9094
Weighted Random Walk	0.8600	0.8289	0.9855	0.5681	0.5230	0.9128	0.4882	0.4154	0.9213
Linear CTDNE	0.9227	0.9012	0.9942	0.5852	0.5356	0.9342	0.5833	0.5149	0.9506
Exponential CTDNE	0.7320	0.6752	0.9620	0.4900	0.4463	0.8889	0.5700	0.4997	0.9375
Temporal Pattern (Freq. 5% Clusters)
Shortest Path Walk	0.9147	0.8929	0.9918	0.6433	0.5980	0.9460	0.5352	0.4624	0.9423
Random Walk	0.9627	0.9521	0.9974	0.7009	0.6687	0.9659	0.6209	0.5535	0.9611
Biased Random Walk	0.7080	0.6537	0.9632	0.6046	0.5733	0.9365	0.4688	0.3957	0.9245
Weighted Random Walk	0.8560	0.8225	0.9815	0.5598	0.5138	0.9209	0.5176	0.4463	0.9312
Linear CTDNE	0.9173	0.8970	0.9907	0.5742	0.5263	0.9205	0.5867	0.5196	0.9516
Exponential CTDNE	0.7467	0.6826	0.9707	0.5193	0.4736	0.8973	0.5448	0.4728	0.9350

Bold highlights the highest score overall for each pattern and dataset. Baseline results are provided for reference in which the italic indicates the highest baseline score