electricsheepafrica/nigerian-banking-personal-loans

Nigerian Personal Loans

Dataset Type: Banking & Finance - Lending
Version: 1.0
License: Apache 2.0
Language: English
Geography: Nigeria

---

Dataset Description

Synthetic Nigerian consumer personal loans with default prediction labels. Includes 3-year loan vintage with realistic employment patterns, credit scoring, and default behavior across Nigerian states.

Use Cases:

• Default probability (PD) modeling
• Loss Given Default (LGD) estimation
• Credit scoring model development
• Loan origination decisioning
• Portfolio risk assessment

---

Dataset Statistics

• Rows: 2,000,000 (pilot: 10,000)
• Columns: 20
• Time Range: 2021-01-01 to 2024-12-31 (applications)
• Observation Date: 2024-12-31
• Default Rate (90d): 8%
• Default Rate (180d): 12%

---

Schema

Column	Type	Description
loan_id	string	Unique loan identifier (LN-########)
customer_id	string	Customer identifier (CUS-########)
application_date	date	Loan application date
disbursement_date	date	Loan disbursement date (if approved)
loan_status	category	rejected, approved, disbursed
principal_ngn	float	Loan principal amount
interest_rate_annual	float	Annual interest rate (decimal)
tenor_months	int	Loan tenor in months
monthly_payment_ngn	float	Monthly payment amount
salary_ngn	float	Monthly salary
debt_to_income_ratio	float	DTI ratio (payment/salary)
credit_score	int	Credit score (300-850 range)
employment_type	category	government, private_sector, self_employed, unemployed
loan_purpose	category	debt_consolidation, business, medical, education, home_improvement, personal
state	category	Nigerian state
salary_detected	bool	True if salary direct deposit detected
dpd_current	int	Days Past Due (current)
dpd_max_90d	int	Maximum DPD in last 90 days
default_90d	bool	LABEL: True if 90+ days past due
default_180d	bool	LABEL: True if 180+ days past due

---

Label Distribution

Default 90d

• Positive: 8%
• Negative: 92%

Default 180d

• Positive: 12%
• Negative: 88%

Note: default_180d is a superset of default_90d

Default by Employment Type

Employment	Default 90d	Default 180d
Government	4%	6%
Private Sector	8%	12%
Self-Employed	15%	22%
Unemployed	25%	35%

---

Data Distributions

Loan Principal

• Distribution: Lognormal (μ=12.5, σ=0.8)
• Range: ₦50,000 - ₦5,000,000
• Median: ~₦300,000
• Mean: ~₦370,000

Interest Rates

• Distribution: Normal (μ=24%, σ=4%)
• Range: 15% - 35% APR
• Mean: 24% (typical Nigerian rates)

Tenor Distribution

Tenor	Percentage
6 months	30%
12 months	40%
18 months	15%
24 months	8%
3 months	5%
36 months	2%

Employment Distribution

• Private Sector: 45%
• Self-Employed: 25%
• Government: 20%
• Unemployed: 10%

Loan Status

• Disbursed: ~60%
• Rejected: ~30%
• Approved (not disbursed): ~10%

---

Default Drivers

High Risk Factors (Increase Default)

• High DTI (>0.5): 8x risk
• Low Credit Score (<550): 6x risk
• Self-Employed: 3.5x risk
• No Salary Detected: 4x risk
• High Interest (>28%): 2.5x risk
• Long Tenor (>24 months): 2x risk

Protective Factors (Decrease Default)

• Government Employee: 0.25x risk
• Salary Detected: 0.4x risk
• High Credit Score (>700): 0.3x risk
• Low DTI (<0.3): 0.5x risk

---

Nigerian Context

Employment Patterns

✅ Government Jobs: More stable, lower default, but fewer
✅ Private Sector: Majority of loans, moderate risk
✅ Self-Employed: High default due to income volatility
✅ Salary Delays: Affects repayment in certain states (Abia, Imo, Plateau)

State-Level Risk

Higher Default States: Abia, Imo, Plateau, Benue, Taraba (salary delays)
Lower Default States: Lagos, Abuja, Rivers (more formal employment)

Loan Purposes

Purpose	Avg Amount	Default Rate
Debt Consolidation	₦500k	12%
Business	₦400k	10%
Medical	₦250k	8%
Education	₦300k	6%
Home Improvement	₦350k	9%
Personal	₦200k	11%

---

Files

personal_loans/
├── README.md
├── nigerian_personal_loans_pilot.parquet (10k, 0.58 MB)
├── nigerian_personal_loans.parquet (2M, ~160 MB) - Coming Soon
├── nigerian_personal_loans.csv (2M, ~320 MB) - Coming Soon
└── personal_loans_sample.csv (100 rows)

---

Usage Example

import pandas as pd
from sklearn.linear_model import LogisticRegression
from sklearn.metrics import roc_auc_score, classification_report

# Load data
df = pd.read_parquet('nigerian_personal_loans.parquet')

# Filter to disbursed loans only
df_disbursed = df[df['loan_status'] == 'disbursed'].copy()

# Feature engineering
features = [
    'debt_to_income_ratio', 'credit_score', 'interest_rate_annual',
    'tenor_months', 'principal_ngn', 'salary_detected'
]

# Encode employment type
df_disbursed['emp_govt'] = (df_disbursed['employment_type'] == 'government').astype(int)
df_disbursed['emp_self'] = (df_disbursed['employment_type'] == 'self_employed').astype(int)
features.extend(['emp_govt', 'emp_self'])

# Train default model
X = df_disbursed[features]
y = df_disbursed['default_90d']

model = LogisticRegression(max_iter=1000)
model.fit(X, y)

# Evaluate
y_pred_proba = model.predict_proba(X)[:, 1]
print(f"ROC-AUC: {roc_auc_score(y, y_pred_proba):.4f}")

---

Limitations

1. Credit Bureau Data: Simplified credit score (real scores require full bureau data)
2. Macroeconomic Factors: Static economic conditions
3. Co-borrowers: Single borrower assumption
4. Collateral: Unsecured loans only
5. Restructuring: Loan modifications not modeled

---

Citation

@dataset{nigerian_personal_loans_2025,
  author = {Electric Sheep Africa},
  title = {Nigerian Personal Loans Dataset},
  year = {2025},
  publisher = {Hugging Face}
}

---

Last Updated: 2025-10-19
Status: ✅ Pilot Validated