Early Rug Pull Warning for BSC Meme Tokens via Multi-Granularity Wash-Trading Pattern Profiling
arXiv:2603.13830v1 Announce Type: new Abstract: 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-AU
arXiv:2603.13830v1 Announce Type: new Abstract: 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.
Executive Summary
This article presents a novel end-to-end framework for detecting rug-pull risks in BSC meme tokens using multi-granularity wash-trading pattern profiling. The framework integrates transaction, address, and flow-level behavioral features across three wash-trading patterns (Self, Matched, Circular) into a supervised warning system. Empirical results demonstrate strong predictive performance using Random Forest (AUC=0.9098, PR-AUC=0.9185, F1=0.7429), with trade-level features as the primary contributor and address-level features providing supplementary value. The model exhibits actionable early-warning capability with an average lead time of 3.81 hours, though its low recall (FN=8) suggests it functions better as a high-precision screener than a comprehensive alarm system. The work contributes a reproducible pipeline, evidence of feature efficacy under weak supervision, and deployment-oriented insights.
Key Points
- ▸ Multi-granularity wash-trading features enhance detection accuracy
- ▸ Random Forest outperforms Logistic Regression on core metrics
- ▸ Lead time of 3.81 hours indicates early warning potential
Merits
Strong Predictive Performance
The framework achieves robust AUC and PR-AUC scores, validating the effectiveness of multi-granularity features in detecting rug-pull risks under weak supervision
Demerits
Limited Recall Capability
The model’s low false negative rate (FN=8) restricts its applicability as an automated alarm engine, limiting scalability for real-time monitoring
Expert Commentary
The paper represents a meaningful advancement in the field of DeFi risk mitigation. The integration of multi-level behavioral signals into a unified risk vector is a sophisticated methodological leap that elevates beyond traditional anomaly detection. The empirical validation of feature performance under weak supervision is particularly noteworthy—it demonstrates that the model’s strength lies not in labeled data abundance but in signal richness and interpretability. The lead-time analysis is pragmatic and adds tangible value for operational deployment. While the low recall rate limits its use as a standalone alarm system, its role as a precision filter is well-justified and complementary to broader monitoring architectures. This work bridges the gap between academic research and real-world applicability in a way that few recent papers have achieved.
Recommendations
- ✓ 1. Extend the model to incorporate on-chain governance and audit trail data to improve recall without sacrificing precision.
- ✓ 2. Publish the labeled dataset and feature modeling code as open resources to enable reproducibility and benchmarking across the DeFi security research community.