NFTruth
AI-Powered NFT Authenticity Detector designed to fight scams with machine learning.
NFTruth is an intelligent security platform that analyzes NFT collections to determine their legitimacy and detect potential scams. Using ensemble machine learning algorithms trained on multi-source data (OpenSea marketplace data, Reddit social sentiment, and Ethereum blockchain metrics), it provides comprehensive risk assessments for NFT collections.
Project Mission
Most NFT fraud detection tools rely exclusively on perceptual image hashing—checking if artwork has been copied. However, modern Web3 scammers often mint original artwork while manufacturing synthetic social volume, wash-trading transactions, and spoofing creator credentials. NFTruth evaluates collections across three distinct operational layers to uncover fraudulent behavior before buyers commit capital.
Tri-Layer Ingestion & Forensic Pipeline
Extracts collection-level verification flags, 24h trading volume consistency, floor price stability, and ownership distribution ratios via OpenSea API.
Uses OAuth 2.0 to stream crypto discussion threads, using VADER sentiment analysis to differentiate organic community engagement from coordinated bot pumping language.
Inspects creator contract deployment age, mint distribution uniformity, transaction velocity, and known wash-trading wallet clusters.
Feeds 20+ engineered signals into a weighted ensemble of four specialized classifiers to compute an explainable risk probability.
Multi-Model Machine Learning Ensemble
Rather than relying on a single monolithic classifier, NFTruth implements an ensemble of four specialized algorithms, each selected for its specific statistical strengths:
- Role: Baseline classification and fast, highly interpretable linear scoring.
- Strength: Establishes calibrated baseline probabilities without overfitting sparse collections.
- Use Case: Provides transparent coefficients for individual risk contributors.
- Role: Feature importance estimation and non-linear signal correlation.
- Strength: Excels at capturing interactions between disparate signal types (e.g., high Twitter mentions paired with zero on-chain liquidity).
- Use Case: Evaluates complex multi-dimensional risk profiles.
- Role: Sequential learning and subtle scam pattern recognition.
- Strength: Iteratively minimizes residual errors to catch sophisticated wash-trading cycles that bypass traditional heuristic thresholds.
- Use Case: Precision detection of organized syndicate fraud.
- Role: High-dimensional decision boundary separation.
- Strength: Effective in sparse, high-dimensional feature spaces constructed from NLP text embeddings and normalized wallet metrics.
- Use Case: Separating borderline legitimate projects from well-disguised scams.
Risk Classification Tiers
Every collection evaluated by NFTruth receives an aggregated risk probability that maps directly to actionable protection tiers:
Forensic Approaches: Image Hashing vs. Multi-Model Ensemble
- Detection Vector: Checks if images are exact duplicates of established collections.
- Vulnerability: Completely blind to original art created for rug-pulls or fake volume syndicates.
- Result: False sense of security for unsuspecting collectors.
- Detection Vector: Evaluates market liquidity, sentiment authenticity, and blockchain provenance simultaneously.
- Coverage: Catches wash trading, bot-driven artificial pumping, and creator wallet cycling regardless of the artwork.
- Result: Transparent, auditable risk score accompanied by concrete warning evidence.
Technology Stack
| Layer | Technology | Purpose |
|---|---|---|
| Logic & Pipeline | Python 3.11 | Core orchestration, data transformation, and CLI scoring |
| Machine Learning | Scikit-learn, XGBoost | Ensemble training, cross-validation, and inference |
| NLP & Sentiment | NLTK, VaderSentiment | Social text classification and bot polarity detection |
| External APIs | OpenSea API, Reddit OAuth, Etherscan | Real-time multi-source data ingestion |
| Data Analysis | Pandas & NumPy | Feature vectorization and matrix manipulation |