This project simulates how fintech lenders in Kenya assess borrower risk using limited but high-frequency data such as income, loan size, and mobile money activity.
# Credit Risk Modelling & Credit Scoring
## Kenya Fintech Simulation
---
## Overview
This project simulates how fintech lenders in Kenya assess borrower risk using limited but high-frequency data such as income, loan size, and mobile money activity.
The goal is to:
- Predict the likelihood of loan default
- Convert predictions into a usable credit score (300–850)
- Demonstrate how machine learning supports lending decisions in fintech environments
---
## Problem Statement
In many fintech lending environments, especially in emerging markets, traditional credit history is often unavailable.
As a result, lenders rely on:
- Behavioral data (e.g., mobile money usage)
- Income patterns
- Loan history
This project builds a model to estimate default risk and translate it into an actionable credit score.
---
## Dataset
A synthetic dataset was created to simulate a Kenyan fintech environment, including:
- **age** — Borrower age
- **monthly_income** — Estimated monthly earnings
- **loan_amount** — Requested loan amount
- **mobile_txn_count** — Mobile money transaction frequency (proxy for financial activity)
- **loan_history** — Number of past loans
- **default** — Target variable (1 = default, 0 = non-default)
---
## Methodology
### 1. Data Simulation
- Generated realistic borrower profiles using statistical distributions
- Introduced relationships between income, loan size, and behavioral variables
### 2. Exploratory Data Analysis (EDA)
- Examined feature distributions
- Checked class balance
- Analyzed correlations between variables
### 3. Feature Engineering
- Derived risk-related features such as:
- Loan-to-income ratio
- Behavioral proxies for financial stability
### 4. Model Development
A **Decision Tree Classifier** was used because:
- It captures non-linear relationships
- It is interpretable
- It is commonly used in risk-based decision systems
### 5. Model Evaluation
The model was evaluated using:
- Precision & Recall
- F1 Score
- ROC-AUC
T …