Traditional Credit Bureau Data
Credit bureaus (Experian, Equifax, TransUnion) maintain comprehensive credit files on 220+ million U.S. consumers. This data forms the foundation of traditional credit scoring.
Credit Report Components
- Personal Information: Name, address history, SSN, date of birth, employment history
- Credit Accounts: All credit cards, loans, mortgages with payment history, balances, limits
- Credit Inquiries: Hard pulls from credit applications (last 2 years)
- Public Records: Bankruptcies, tax liens, civil judgments, foreclosures
- Collections: Accounts sent to collection agencies
Key Bureau Data Features
- Number of accounts (total, open, closed)
- Account ages (oldest, newest, average)
- Credit limits and balances
- Utilization ratios (overall and per-account)
- Payment history (on-time, late 30/60/90+ days)
- Derogatory marks count and severity
- Hard inquiries (6 months, 12 months, 24 months)
- Mix of credit types
Alternative Data Sources
Alternative data enables scoring for credit-invisible consumers and improves accuracy for all applicants.
1. Utility Payments
- Electric, gas, water bills
- Cable, internet, phone services
- Demonstrates payment responsibility
- Providers: Experian Boost, eCredable, PRBC
2. Rent Payment History
- Monthly housing payments
- Typically largest recurring expense
- Strong predictor of loan repayment
- Providers: RentTrack, ClearNow, PayYourRent
3. Open Banking / Bank Transactions
- Account balances and cash flow
- Income verification and stability
- Spending patterns and categories
- Overdrafts and NSF fees
- Savings behavior
- Providers: Plaid, Yodlee, Finicity, MX
4. Telecom and Mobile
- Mobile phone payment history
- Account tenure and upgrades
- Plan type and payment method
5. Employment and Income
- Employment verification (The Work Number, Truework)
- Income amount and stability
- Job tenure and industry
6. Education Credentials
- Degree level and field of study
- Educational institution quality
- Completion status
- Correlates with income potential
Feature Engineering Techniques
Aggregation Features
- Total number of accounts (all types, by type)
- Sum of all credit limits
- Average account age
- Total outstanding balance
Ratio Features
- Debt-to-income (DTI) ratio
- Credit utilization (overall and per card)
- Payment-to-income ratio
- Revolving vs. installment debt ratio
Trend Features
- Balance change over 3/6/12 months
- Utilization trend (increasing/decreasing)
- Number of new accounts (3/6/12 months)
- Credit limit changes
Behavioral Features
- Payment timing (early, on-time, late)
- Minimum payment vs. full balance
- Cash advance usage
- Credit-seeking behavior (inquiries)
Transaction-Based Features (from Open Banking)
- Monthly income (median, variance)
- Recurring expenses identification
- Savings rate and consistency
- Gambling or risky spending patterns
- Financial stress indicators (overdrafts, payday loans)
- Positive indicators (investments, insurance payments)
Advanced Feature Engineering
Domain Expertise + ML: Combine financial domain knowledge with automated feature generation. Modern systems can create 500-1000+ features automatically while maintaining interpretability through feature importance analysis.
Data Quality and Preparation
Missing Data Handling
- Imputation: Fill missing values with median/mode/model-based estimates
- Indicator Variables: Create "missing" flags to capture information
- Model-Based: Use ML to predict missing values
Outlier Treatment
- Winsorization (cap extreme values at percentiles)
- Log transformation for skewed distributions
- Binning for extreme ranges
Data Normalization
- Standardization (mean=0, std=1) for neural networks
- Min-max scaling for bounded ranges
- Rank-based normalization for non-parametric approaches
Privacy and Compliance
Critical Requirements
- Permissible Purpose: FCRA requires legitimate use of credit data
- Consumer Consent: Explicit permission for alternative data access
- Data Minimization: Collect only necessary information
- Prohibited Factors: Never use race, gender, religion, national origin
- Proxy Detection: Monitor for variables that correlate with protected classes
- Right to Explanation: Provide adverse action reasons from actual features used
Next chapter covers implementing these data pipelines and scoring systems in production environments.