Organizations: School of Big Data, Baoshan University, Baoshan 678000, Yunnan, China · School of Information Science and Electrical Engineering, Shandong Jiaotong University, Jinan, Shandong, China
The high-frequency issuance and short-cycle speculation of meme tokens in decentralized finance (DeFi) have significantly amplified rug-pull risk. Existing approaches still struggle to provide stable early warning under scarce anomalies, incomplete labels, and limited interpretability. To address this issue, an end-to-end warning framework is proposed for BSC meme tokens, consisting of four stages: dataset construction and labeling, wash-trading pattern feature modeling, risk prediction, and error analysis. Methodologically, 12 token-level behavioral features are constructed based on three wash-trading patterns (Self, Matched, and Circular), unifying transaction-, address-, and flow-level signals into risk vectors. Supervised models are then employed to output warning scores and alert decisions. Under the current setting (7 tokens, 33,242 records), Random Forest outperforms Logistic Regression on core metrics, achieving AUC=0.9098, PR-AUC=0.9185, and F1=0.7429. Ablation results show that trade-level features are the primary performance driver (Delta PR-AUC=-0.1843 when removed), while address-level features provide stable complementary gain (Delta PR-AUC=-0.0573). The model also demonstrates actionable early-warning potential for a subset of samples, with a mean Lead Time (v1) of 3.8133 hours. The error profile (FP=1, FN=8) indicates that the current system is better positioned as a high-precision screener rather than a high-recall automatic alarm engine. The main contributions are threefold: an executable and reproducible rug-pull warning pipeline, empirical validation of multi-granularity wash-trading features under weak supervision, and deployment-oriented evidence through lead-time and error-bound analysis.
Figures & tables
Figure 1: BSC Meme Token Rug Pull Early-warning Framework. The overall pipeline contains four stages (E1–E4): data construction and labeling, wash-trading pattern profiling, early-warning modeling, and ablation/error analysis.
Item
Value
Number of tokens
7
Total transfer records
33,242
Per-token cap (records)
5,000
Minimum records per token
3,256
Median records per token
5,000
Maximum records per token
5,000
Table 1: Summary statistics of the curated BSC meme-token dataset after preprocessing and deduplication.
Token Address
Records
0x595d…9cc
4,994
0x4fa7…3bf
3,256
0x5440…1b0
5,000
0xe6df…528
5,000
0x8d0d…0d
4,992
0xe0a2…2e
5,000
Table 2: Token-level record counts after preprocessing.
Figure 2: Data Collection and Labeling Pipeline. The workflow includes export, token-wise merging, deduplication, normalization, window capping, rule-based labeling, and quality-control checks.
Figure 3: Wash-trading Pattern Profiling and Risk Vector Construction. Transaction-, address-, and flow-level signals are scored into Self/Matched/Circular components and then aggregated into token-level risk vectors.
Feature
Category
Definition
Interpretation
tx_count
Activity
Number of transfer records
Overall trading activity
unique_from
Addr. diversity
Unique sender addresses
Outgoing-side participation
unique_to
Addr. diversity
Unique receiver addresses
Incoming-side participation
from_to_ratio
Struct. balance
unique_from/unique_to
Sender/receiver asymmetry
avg_quantity
Amount stats
Mean transfer quantity
Average transfer size
median_quantity
Amount stats
Median transfer quantity
Typical transfer size
Table 3: Definition of token-level behavioral features used in early rug-pull warning.
Figure 4: Early-warning Modeling and Deployment Flow. The upper lane shows model training and evaluation. The lower lane shows inference and alert triggering, with lead-time estimation as deployment output.
Model
Accuracy
Precision
Recall
F1
AUC
PR-AUC
Lead Time (h)
Logistic Regression
0.6500
0.7059
0.5714
0.6316
0.7243
0.7397
3.8133
Random Forest
0.7750
0.9286
0.6190
0.7429
0.9098
0.9185
3.8133
Table 4: Main early-warning performance comparison. Best values are boldfaced.
Figure 5: Main Performance Comparison between Logistic Regression and Random Forest. (a) Classification metrics. (b) Ranking metrics.
Figure 6: Performance vs. Lead Time Trade-off (v1). AUC and PR-AUC are shown together with lead time.
Setting
#Feat.
Prec.
Recall
F1
AUC
PR-AUC
Δ PR-AUC
Full model
12
0.9286
0.6190
0.7429
0.9098
0.9185
+0.0000
w/o trade
6
0.7647
0.6190
0.6842
0.7168
0.7342
−0.1843
w/o address
7
0.7000
0.6667
0.6829
0.7920
0.8612
−0.0573
w/o contract
11
0.9231
0.5714
0.7059
0.9173
0.9262
+0.0077
Table 5: Feature-group ablation of the Random Forest model. Δ PR-AUC is computed against the full model.
Split
FP
FN
Observation
Test split
1
8
Conservative thresholding with limited recall
Table 6: Error profile of the Random Forest model on the test split.
Figure 7: Case Timeline from Warning to Rug Pull. Representative cases visualize warning-to-event intervals and demonstrate heterogeneity in risk-evolution speed.