Illicit Transaction Detection on Elliptic Bitcoin Louvain partitioning (test)
90PrecisionFedSage+
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| FedSage+Federation approach=Federated, Partitioning method=Louvain, Number of training rounds=50, Backbone architecture=2-layer GraphSAGE2026.01 | 90 | 48 | 45.3 | |
| Local GNNFederation approach=No Federation, Partitioning method=Louvain, Number of training rounds=50, Backbone architecture=2-layer GraphSAGE2026.01 | 74 | 35 | 38.9 | |
| FedGraph-VASPFederation approach=Federated, Partitioning method=Louvain, Number of training rounds=50, Backbone architecture=2-layer GraphSAGE2026.01 | 64 | 43 | 50.8 | |
| FedAvgFederation approach=Federated, Partitioning method=Louvain, Number of training rounds=50, Backbone architecture=2-layer GraphSAGE2026.01 | 58 | 45 | 49.9 |