Table 4 Ablation results of AdaAdvSAGE module.

From: HMOA-GNN: adaptive adversarial GraphSAGE with hierarchical hybrid sampling and metric-optimized graph construction for credit card fraud detection

Dataset

Algorithm

Evaluation Metrics

Accuracy

Precision

Recall

F1-score

AUC

European

Baseline

0.9745

0.0000

0.0000

0.0000

0.5000

Baseline w/ AdaAdvSAGE

0.9745(−)

0.0000(−)

0.0000(−)

0.0000(−)

0.5000(−)

HMOA-GNN

0.9965

0.9400

0.9216

0.9307

0.9975

HMOA-GNN w/o AdaAdvSAGE

0.9965(−)

1.0000(\(\uparrow\))

0.8824(\(\downarrow\))

0.9278(\(\downarrow\))

0.9507(\(\downarrow\))

IEEE-CIS

Baseline

0.9815

0.8718

0.5152

0.6476

0.9099

Baseline w/ AdaAdvSAGE

0.9900(\(\uparrow\))

0.9259(\(\uparrow\))

0.7576(\(\uparrow\))

0.8333(\(\uparrow\))

0.9580(\(\uparrow\))

HMOA-GNN

0.9805

0.6627

0.8333

0.7383

0.9445

HMOA-GNN w/o AdaAdvSAGE

0.1450(\(\downarrow\))

0.0361(\(\downarrow\))

1.0000(\(\uparrow\))

0.0696(\(\downarrow\))

0.5452(\(\downarrow\))

Simulated

Baseline

0.9685

0.0000

0.0000

0.0000

0.9386

Baseline w/ AdaAdvSAGE

0.9795(\(\uparrow\))

0.7059(\(\uparrow\))

0.5806(\(\uparrow\))

0.6372(\(\uparrow\))

0.9032(\(\downarrow\))

HMOA-GNN

0.9815

0.7049

0.6935

0.6992

0.9177

HMOA-GNN w/o AdaAdvSAGE

0.9830(\(\uparrow\))

0.7500(\(\uparrow\))

0.6774(\(\downarrow\))

0.7119(\(\uparrow\))

0.8931(\(\downarrow\))

Back to article page