Catching the Rug: Early Prediction of Fraudulent Memecoins on Solana via Machine Learning
Using XGBoost with first 5-minute trading data, predict Solana memecoin rug pulls within 1 hour.
Key Findings
Methodology
This study assembled a large-scale dataset of 6.4 million Solana memecoins over 7 months, extracting features based on trading volume and liquidity dynamics. Using models like XGBoost, the research conducted time-series rolling validation to prevent data leakage and assessed cross-platform transferability between PumpFun and Raydium. The approach relies solely on early trading data, enabling prediction of rug pull risks within an hour, significantly outperforming baseline models. The methodology emphasizes feature importance analysis and multi-source data fusion to address domain shifts, demonstrating robustness in high-throughput blockchain environments.
Key Results
- XGBoost achieved an F1-score of 0.78, MCC of 0.65, and AUCPRC of 0.81 in predicting rug pulls within 1 hour using only the first 5 minutes of trading data, outperforming Random Forest and MLP.
- Cross-platform experiments showed a performance increase of approximately 12% after integrating data from PumpFun and Raydium, confirming the effectiveness of multi-source fusion.
- The model maintained high accuracy even in high liquidity scenarios, validating the practicality of short-horizon early warning in real-time blockchain environments.
Significance
This work advances the field by providing a scalable, real-time risk detection framework that does not depend on smart contract code analysis. It addresses critical challenges in DeFi security, such as data imbalance, rapid market microstructure changes, and platform-specific distribution shifts. The approach enhances investor protection and market stability, paving the way for more resilient decentralized finance ecosystems on high-throughput chains like Solana.
Technical Contribution
The research introduces a novel liquidity-based early warning framework utilizing gradient boosting models, with a focus on feature engineering from transaction and liquidity data. It employs time-aware cross-validation and multi-source data fusion to improve generalization across platforms. The methodology avoids reliance on source code, making it adaptable and computationally efficient for high-frequency environments. These innovations collectively push forward the frontier of chain-level fraud detection.
Novelty
This is the first large-scale (6.4 million tokens) study on Solana memecoin rug pull prediction using only trading behavior features within a short 5-minute window. It emphasizes cross-platform transferability and multi-source data fusion, addressing the domain shift problem that limits previous models. The focus on liquidity dynamics rather than smart contract code marks a significant departure from traditional approaches, offering a practical, scalable solution for real-time risk management.
Limitations
- The model's reliance on trading activity and liquidity indicators may limit effectiveness during extreme market volatility or novel scam strategies. Additional microstructure features could be incorporated for robustness.
- Short time window (5 minutes) may reduce accuracy in low liquidity or highly volatile conditions, necessitating adaptive window strategies.
- High computational cost for training and inference, especially at large scale, requires further optimization for deployment in production environments.
Future Work
Future efforts will explore integrating deep learning models like Transformers for richer feature representation, incorporating social media sentiment analysis, and extending the framework to other high-throughput chains such as Binance Smart Chain and Polygon. Enhancing model interpretability and real-time deployment efficiency will also be prioritized to facilitate practical adoption.
AI Executive Summary
The explosive growth of memecoins on blockchain platforms like Solana has brought both innovation and risk. While memecoins attract speculative traders with their viral appeal, they are also prime targets for fraudulent schemes such as rug pulls, which can wipe out investor funds rapidly. Traditional detection methods often rely on analyzing smart contract code, but this approach faces limitations in high-frequency trading environments and with obfuscated or malicious code. To address this, our research develops a novel early warning system based solely on transaction behavior and liquidity metrics.
Leveraging a dataset of 6.4 million tokens collected over seven months, we engineered features capturing trading volume, price dynamics, and liquidity fluctuations. Using gradient boosting models like XGBoost, trained with a time-aware rolling validation strategy, we achieved high predictive accuracy within a 5-minute window to forecast rug pulls occurring within one hour. The models demonstrated strong performance both within single platforms and across different decentralized exchanges, such as PumpFun and Raydium, especially when employing multi-source data fusion to mitigate domain shifts.
This approach signifies a paradigm shift from traditional smart contract analysis to behavior-based risk detection, offering a practical, scalable solution for real-time security in high-throughput blockchain ecosystems. The findings highlight the importance of liquidity dynamics as early indicators of fraud, providing valuable tools for investors and platform operators to prevent losses and maintain trust.
Despite promising results, limitations remain, including challenges in low liquidity scenarios and computational costs. Future work aims to incorporate deep learning techniques, social media signals, and cross-chain generalization, striving toward a comprehensive, intelligent DeFi risk management framework that adapts to evolving scam tactics and market conditions.
Deep Dive
Abstract
The rapid proliferation of memecoins on blockchain platforms has increased the risk of fraudulent activities, particularly rug pulls. While previous studies have focused on Ethereum-based tokens, this paper shifts the spotlight to Solana, the leading blockchain for memecoins by trading volume and token count. Unlike Ethereum, where rug pulls often exploit smart contract backdoors, Solana memecoin rug pulls are predominantly driven by liquidity manipulation and social dynamics. This research pioneers large-scale rug pull early detection in the Solana ecosystem by assembling a dataset of 6.4 million tokens over 7 months. Market analysis reveals that a vast majority of these memecoins exhibit rug pull characteristics within one hour of launch, highlighting the urgency of short-horizon prediction. Despite the absence of code-level features, we demonstrate that classic machine learning models, particularly Gradient Boosting (XGBoost), achieve robust performance in detecting potential rug pulls using only the first 5 minutes of trading data. Furthermore, we evaluate cross-platform generalization between PumpFun and Raydium, revealing that multi-source data fusion significantly mitigates domain shift and improves detection reliability. This study advances the understanding of DeFi fraud on high-throughput chains and provides a practical framework for protecting investors.