temporal-twins-code / results /paper_suite_summary.md
temporal-twins-anon's picture
Add anonymous Temporal Twins code release
a3682cf verified

Final Paper Suite Summary

Rows: 20

Dataset and Audit

Benchmark Matched Pairs Positives Negatives Fraud Users Benign Users Templates Positive Rate Benign Motif Hit static_agg_auc Txn AUC Idx AUC Prefix AUC Time AUC Account AUC Active AUC
easy 2222.2000 ± 128.4434 2222.2000 ± 128.4434 2222.2000 ± 128.4434 304.6000 ± 23.2551 304.6000 ± 23.2551 300.2000 ± 24.2425 0.5000 ± 0.0000 0.0000 ± 0.0000 0.5000 ± 0.0000 0.5000 ± 0.0000 0.5000 ± 0.0000 0.5000 ± 0.0000 0.5000 ± 0.0000 0.5000 ± 0.0000 0.5000 ± 0.0000
hard 2317.6000 ± 21.9727 2317.6000 ± 21.9727 2317.6000 ± 21.9727 315.0000 ± 0.0000 315.0000 ± 0.0000 315.0000 ± 0.0000 0.5000 ± 0.0000 0.0000 ± 0.0000 0.5000 ± 0.0000 0.5000 ± 0.0000 0.5000 ± 0.0000 0.5000 ± 0.0000 0.5000 ± 0.0000 0.5000 ± 0.0000 0.5000 ± 0.0000
medium 2356.6000 ± 18.0361 2356.6000 ± 18.0361 2356.6000 ± 18.0361 263.0000 ± 0.0000 263.0000 ± 0.0000 263.0000 ± 0.0000 0.5000 ± 0.0000 0.0000 ± 0.0000 0.5000 ± 0.0000 0.5000 ± 0.0000 0.5000 ± 0.0000 0.5000 ± 0.0000 0.5000 ± 0.0000 0.5000 ± 0.0000 0.5000 ± 0.0000
oracle_calib 2606.6000 ± 454.3449 2606.6000 ± 454.3449 2606.6000 ± 454.3449 409.8000 ± 86.5402 409.8000 ± 86.5402 409.8000 ± 86.5402 0.5000 ± 0.0000 0.0000 ± 0.0000 0.5000 ± 0.0000 0.5000 ± 0.0000 0.5000 ± 0.0000 0.5000 ± 0.0000 0.5000 ± 0.0000 0.5000 ± 0.0000 0.5000 ± 0.0000

Probes and Models

Benchmark Primary Probe Secondary Probe XGB ROC/PR StaticGNN ROC/PR SeqGRU Clean ROC/PR SeqGRU Shuf ROC/PR SeqGRU Delta
easy MotifProbe 1.0000 ± 0.0000 RawMotifProbe 0.9983 ± 0.0011 0.5000 ± 0.0000 / 0.5000 ± 0.0000 0.4946 ± 0.0128 / 0.5001 ± 0.0099 1.0000 ± 0.0000 / 1.0000 ± 0.0000 0.4997 ± 0.0096 / 0.5008 ± 0.0077 -0.5003 ± 0.0096
hard MotifProbe 0.5790 ± 0.0045 RawMotifProbe 0.5910 ± 0.0105 0.5000 ± 0.0000 / 0.5000 ± 0.0000 0.5026 ± 0.0198 / 0.5081 ± 0.0128 0.6876 ± 0.0128 / 0.7058 ± 0.0250 0.4994 ± 0.0033 / 0.4981 ± 0.0031 -0.1883 ± 0.0111
medium MotifProbe 0.6374 ± 0.0069 RawMotifProbe 0.6482 ± 0.0058 0.5000 ± 0.0000 / 0.5000 ± 0.0000 0.4922 ± 0.0203 / 0.4956 ± 0.0231 0.8391 ± 0.0174 / 0.8520 ± 0.0110 0.5053 ± 0.0071 / 0.5054 ± 0.0078 -0.3337 ± 0.0191
oracle_calib AuditOracle 1.0000 ± 0.0000 RawMotifOracle 1.0000 ± 0.0000 0.5000 ± 0.0000 / 0.5000 ± 0.0000 0.5222 ± 0.0235 / 0.5141 ± 0.0202 1.0000 ± 0.0000 / 1.0000 ± 0.0000 0.4968 ± 0.0043 / 0.5008 ± 0.0057 -0.5032 ± 0.0043

Temporal GNNs

Benchmark TGN ROC/PR/Delta TGAT ROC/PR/Delta DyRep ROC/PR/Delta JODIE ROC/PR/Delta
easy 0.5660 ± 0.0123 / 0.5566 ± 0.0107 / -0.0668 ± 0.0126 0.5200 ± 0.0119 / 0.5177 ± 0.0086 / -0.0294 ± 0.0244 0.5320 ± 0.0090 / 0.5245 ± 0.0100 / -0.0306 ± 0.0150 0.5282 ± 0.0085 / 0.5210 ± 0.0079 / -0.0420 ± 0.0231
hard 0.5060 ± 0.0081 / 0.5069 ± 0.0062 / -0.0006 ± 0.0131 0.5157 ± 0.0135 / 0.5133 ± 0.0163 / -0.0145 ± 0.0173 0.5229 ± 0.0050 / 0.5172 ± 0.0072 / -0.0171 ± 0.0147 0.5137 ± 0.0085 / 0.5087 ± 0.0076 / -0.0231 ± 0.0139
medium 0.5177 ± 0.0063 / 0.5133 ± 0.0034 / -0.0124 ± 0.0094 0.5199 ± 0.0146 / 0.5121 ± 0.0105 / -0.0334 ± 0.0211 0.5285 ± 0.0246 / 0.5252 ± 0.0144 / -0.0290 ± 0.0181 0.5298 ± 0.0057 / 0.5235 ± 0.0103 / -0.0299 ± 0.0061
oracle_calib 0.6270 ± 0.0103 / 0.6107 ± 0.0144 / -0.1288 ± 0.0173 0.6327 ± 0.0465 / 0.6104 ± 0.0381 / -0.1403 ± 0.0654 0.6135 ± 0.0481 / 0.6010 ± 0.0400 / -0.1203 ± 0.0600 0.5903 ± 0.0170 / 0.5720 ± 0.0146 / -0.0947 ± 0.0227

Runtime

Benchmark Run Time (sec) StaticGNN Eval Time (sec)
easy 1345.9431 ± 135.9206 198.4129 ± 45.3445
hard 2613.7203 ± 231.4288 272.9988 ± 70.8384
medium 2181.9480 ± 228.0777 184.2332 ± 57.5306
oracle_calib 1136.5994 ± 283.1015 170.9515 ± 34.9402

Failed Gate Checks

Benchmark Seed Gate Pass Volume Failures Hard Gate Failures Advisory Failures
easy 0 1 - - TGN ROC-AUC: 0.5835 (>=0.75)
easy 1 1 - - TGN ROC-AUC: 0.5495 (>=0.75)
easy 2 1 - - TGN ROC-AUC: 0.5638 (>=0.75)
easy 3 1 - - TGN ROC-AUC: 0.5693 (>=0.75)
easy 4 1 - - TGN ROC-AUC: 0.5639 (>=0.75)
hard 0 0 - MotifProbe ROC-AUC: 0.5838 (>=0.99) MotifProbe pair-sep: 0.1676 (>=0.99)
hard 1 0 - MotifProbe ROC-AUC: 0.5771 (>=0.99) MotifProbe pair-sep: 0.1543 (>=0.99)
hard 2 0 - MotifProbe ROC-AUC: 0.5824 (>=0.99) MotifProbe pair-sep: 0.1648 (>=0.99)
hard 3 0 - MotifProbe ROC-AUC: 0.5724 (>=0.99) MotifProbe pair-sep: 0.1448 (>=0.99)
hard 4 0 - MotifProbe ROC-AUC: 0.5790 (>=0.99) MotifProbe pair-sep: 0.1581 (>=0.99)
medium 0 0 - MotifProbe ROC-AUC: 0.6463 (>=0.99) MotifProbe pair-sep: 0.2926 (>=0.99)
medium 1 0 - MotifProbe ROC-AUC: 0.6303 (>=0.99) MotifProbe pair-sep: 0.2606 (>=0.99)
medium 2 0 - MotifProbe ROC-AUC: 0.6423 (>=0.99) MotifProbe pair-sep: 0.2846 (>=0.99)
medium 3 0 - MotifProbe ROC-AUC: 0.6314 (>=0.99) MotifProbe pair-sep: 0.2629 (>=0.99)
medium 4 0 - MotifProbe ROC-AUC: 0.6366 (>=0.99) MotifProbe pair-sep: 0.2731 (>=0.99)
oracle_calib 0 1 nan nan TGN ROC-AUC: 0.6135 (>=0.75)
oracle_calib 1 1 - - TGN ROC-AUC: 0.6250 (>=0.75)
oracle_calib 2 1 - - TGN ROC-AUC: 0.6353 (>=0.75)
oracle_calib 3 1 - - TGN ROC-AUC: 0.6223 (>=0.75)
oracle_calib 4 1 - - TGN ROC-AUC: 0.6391 (>=0.75)