ziadatalabs/FreeSyntheticCreditRisk50M

Free Synthetic Credit Risk / Loan Applications (50M)

A free, fully synthetic dataset of 50,000,000 loan application records, built for developers and researchers working on credit-risk models, underwriting logic, or fintech prototypes who need realistic-looking application data without touching any real applicant information.

Every value in this dataset is artificially generated. No real applicants, no real credit histories, no real PII. Approval outcomes and pricing are deliberately correlated with the underlying risk signals — this isn't random noise, it's built so a model trained or tested against it has to actually learn the relationship between risk factors and outcomes, not just pattern-match on nothing.

Schema

Column	Type	Description
application_id	string	Unique identifier for the application
applicant_age	int	Synthetic applicant age
annual_income	float	Synthetic annual income (USD)
employment_status	string	One of: employed, self_employed, unemployed, retired
credit_score	int	Synthetic credit score (300–850 range)
loan_amount	float	Requested loan amount (USD)
loan_purpose	string	One of: debt_consolidation, home_improvement, auto, education, business, other
debt_to_income_ratio	float	Synthetic DTI ratio
existing_loans_count	int	Number of other open loans
application_date	string	ISO date (YYYY-MM-DD), spread across ~3 years
loan_status	string	One of: approved, denied, pending
interest_rate	float	Offered rate (%), only populated when approved — null otherwise

How the correlations work

• Approval likelihood rises with credit_score and falls with debt_to_income_ratio — higher-risk profiles are denied more often.
• interest_rate is only set for approved applications and is priced inversely to credit_score — lower scores get higher rates, matching real risk-based pricing.
• existing_loans_count trends slightly higher for lower credit scores.
• A small slice of applications (~5%) are pending regardless of risk profile, matching real-world review queues.

Format

Single Parquet file, Snappy compression, ~1.16 GB, 50,000,000 rows.

Quick start

import pandas as pd
df = pd.read_parquet("credit_risk_50M.parquet")

from datasets import load_dataset
ds = load_dataset("ziadatalabs/FreeSyntheticCreditRisk50M")

import duckdb
con = duckdb.connect()
con.sql("SELECT * FROM 'credit_risk_50M.parquet' LIMIT 10").show()

Notes

• All applicants, incomes, and credit histories are fabricated — no real individuals represented.
• interest_rate is null for denied and pending applications — don't treat a missing rate as zero.
• Use loan_status and interest_rate together as your prediction targets; the underlying risk signals (credit_score, debt_to_income_ratio, existing_loans_count) are what drive them.

License & Usage

Released under CC BY-NC 4.0 — free for personal, research, and educational use, with attribution. Not licensed for commercial use.

No real people, applications, or financial institutions are represented in this data. It is entirely synthetic.

---

Created by Zia Data Labs. Questions or feedback: zia.data.team@protonmail.com