Blockchain Wallet Credit Score Analysis

Overview

Using raw transaction-level data from lending platforms on the blockchain, I:

• Engineered meaningful behavioral features for each wallet
• Normalized these features
• Scored wallets using a heuristic-based approach
• Output scores and analyze behavior patterns

Methodology

I used heuristic scoring based on engineered behavioral features. Each wallet is scored between 0 and 1000 based on how reliably and responsibly it interacted with the lending protocol.

Processing Flow

1. Load & Parse: Read the raw transaction JSON file, convert timestamps, and extract the action and amount fields.
2. Group by Wallet: Aggregate transactions by userWallet to compute wallet-level features.
3. Feature Engineering: Extract behavioral metrics such as number of deposits, borrowing behavior, liquidations, action diversity, etc.
4. Normalize Features: Apply MinMaxScaler to ensure features are on the same scale.
5. Score Computation: Multiply normalized features by defined weights and scale to range [0, 1000].
6. Export Results: Save credit scores as a CSV and generate a score distribution plot.

Features Engineered

• repay_ratio: total repaid / total borrowed
• borrow_to_deposit_ratio: leverage measure
• num_liquidations: penalizes risky users
• avg_time_between_actions: irregularity suggests organic use
• distinct_days_active: higher = long-term user
• action_entropy: measures diversity of actions

These are engineered by grouping the transaction data per wallet and calculating metrics that indicate responsible DeFi behavior

Technologies Used

• Python: as the programming language.
• Pandas: to extract the raw JSON data and access it when needed.
• NumPy: to define heuristic weights
• Scikit-learn: to normalize, apply weighted scores and scale to the 0-1000 range.
• Matplotlib: to plot the histogram bar chart.
• ChatGPT: as my coding companion.

Outcome

• A CSV file with all the credit scores
• And a histogram plot showing the distribution of scores across all wallets.