Ethereum Fraud Detection

Milestones

Part 1. Data Prep (Saksham)

• Preparation:
  • Focus on current database for now
• Feature Evaluation:
  • PCA → find the good features (Kenny)
  • Heatmap (Akshay)
• Balanced Data / Training Data prep:
  • 3-Way cross validation
  • SMOTE (Pratik)

Part 2. Model Choice

• Supervised:
  • RF, DT (Akshay)
• Unsupervised:
  • GMM, DBSCAN, Multiple Component Multivariate GMM (Pratik)

Part 3. Model Evaluation (Kenny)

• ROC, PRC (F1)

• convert string to numeric (S)

• normalized the data ()

• features explaination

Notes:

• Etherscan column verification: confirm data with etherscan (Currently all values are 0)
  • ERC20 min val sent contract
  • ERC20 max val sent contract
  • ERC20 avg val sent contract
  • ERC20 avg time between sent tnx
  • ERC20 avg time between rec tnx
  • ERC20 avg time between rec 2 tnx
  • ERC20 avg time between contract tnx