The SentinelX pipeline has six stages: input processing (Etherscan source retrieval or raw Solidity), feature extraction across 53 security signals, rule-based detection for known danger patterns, XGBoost machine-learning classification, SHAP explainability, and a final risk report scored from 0 to 100.
The model was trained on 2,400+ labeled contracts with an 80/20 train-test split and achieved 96.15% test accuracy. Feature extraction is deterministic regex and AST-driven analysis of source code; the ML layer classifies, and rules override for critical findings. Full details, feature taxonomy, and evaluation metrics are published on this page.
Related pages
SentinelX provides automated security analysis for informational purposes only. Results are not financial advice or guarantees of safety.