Home

Data, EDA & Preprocessing

What you'll see here

Which loan traits separate defaulters from repayers, and how the data was prepared for modeling.

Rows × features
93,174 × 24
Duplicate IDs
0
Columns with gaps
9
Engineered features
13
Default rate by loan grade

Default risk climbs steadily from grade A to G.

Default rate by weeks of EMI paid

Repayment history is the strongest single signal.

Default rate by interest-rate quintile
Default rate by interest received (quintile)
Default rate by loan term
Default rate by income verification
Default rate by home ownership
Default rate by job experience
Missing values and treatment
  • job_experience4,702
  • annual_income1
  • delinq_2yrs2
  • public_records2
  • total_acc2
  • last_week_pay1,924
  • total_current_balance7,386
  • total_revolving_limit7,386
  • _lwp_band1,924

Tree models handle gaps natively; a missing-flag feature preserves the signal that a value was absent. ID was dropped.

Engineered features
  • Loan term in months
  • Job experience as years + missing flag
  • Missing flags for balance and weeks paid
  • Sub-grade as an ordinal number
  • Interest received ÷ loan amount
  • Weeks paid ÷ term (repayment progress)
  • Expected interest to date
  • Received − expected interest gap
  • Interest received per week paid
  • Revolving utilisation
  • Loan ÷ income
  • Monthly instalment
  • Instalment ÷ income

None use the target, so there is no leakage. Validation: 5-fold stratified CV, seed 42.