Credit Scoring Models

Credit scoring is central to Trumnix's lending operations. The AI models evaluate both on-chain data and off-chain data to generate dynamic and accurate credit scores.

Data Inputs:

1. On-Chain Data:

   • Wallet Activity: Number and frequency of transactions.

   • Transaction History: Historical loan repayments and participation in DeFi protocols.

   • Collateral Holdings: Value and diversity of digital assets held.

2. Off-Chain Data:

   • Utility Bills: Payment consistency and historical trends.

   • Social Media Interactions: Verified profiles and activity to assess borrower credibility.

   • Employment and Income Data: Provided via secure integrations oracles.

Formula for Credit Scoring:

The credit score (CSCSCS) is computed using a weighted scoring system that combines on-chain (OCOCOC) and off-chain (OFOFOF) data, processed through a machine learning model:

CS=σ(w1⋅OC+w2⋅OF+b)CS = \sigma \left( w_1 \cdot OC + w_2 \cdot OF + b \right)CS=σ(w1​⋅OC+w2​⋅OF+b)

Where:

• w1w_1w1​ and w2w_2w2​: Weights for on-chain and off-chain data, learned during model training.

• bbb: Bias term for normalization.

• σ\sigmaσ: Activation function (e.g., sigmoid) to bound the score between 0 and 1.

Protocols:

1. Data Normalization Protocol: Ensures all inputs are standardized (e.g., Min-Max Scaling) before entering the model to prevent bias.

2. Privacy Protocols: Off-chain data is accessed only with user consent, and personal information is anonymized using homomorphic encryption techniques.