📊 Credit Risk EDA Case Study

Exploratory Data Analysis to identify patterns in loan application data that drive credit default risk.

---

🧠 Objective

Financial institutions face significant losses when loans are approved for applicants who are unable to repay. This case study uses Exploratory Data Analysis (EDA) to uncover the key factors that distinguish defaulters (Target = 1) from non-defaulters (Target = 0) — enabling smarter, data-driven lending decisions.

---

📁 Dataset

Two datasets from a real-world lending institution:

File	Description
application_data.csv	Current loan applications with client demographics, financials, and repayment history
previous_application.csv	Historical loan applications and their outcomes for the same clients

Target Variable: TARGET — 1 = client had payment difficulties, 0 = all payments on time

---

🔍 Analysis Workflow

1. Data Cleaning

• Identified and dropped columns with >13% missing values
• Applied percentile clipping (5th–95th) to handle outliers in skewed numeric columns such as AMT_ANNUITY, AMT_GOODS_PRICE, and social circle observation counts
• Imputed remaining nulls using mean (numeric) and mode/domain logic (categorical)

2. Feature Engineering

• LOAN_PERIOD — derived as AMT_CREDIT / AMT_ANNUITY to estimate loan duration
• AGE — converted from DAYS_BIRTH (negative days) to positive years
• AGE_BIN — bucketed age into decade ranges (11–20, 21–30, ..., 61–70) for cohort analysis

3. Univariate Analysis

Compared distributions across Target = 0 and Target = 1 for:

• Income type (NAME_INCOME_TYPE)
• Education level (NAME_EDUCATION_TYPE)
• Family status (NAME_FAMILY_STATUS)
• Housing situation (NAME_HOUSING_TYPE)
• Contract type (NAME_CONTRACT_TYPE)
• Day of week for application (WEEKDAY_APPR_PROCESS_START)

4. Bivariate & Correlation Analysis

• Pearson correlation heatmaps for both defaulter and non-defaulter segments
• Scatter plots exploring relationships between credit amount, goods price, and annuity
• Box plots comparing numerical variables across risk segments

5. Previous Application Analysis

• Univariate analysis of contract type, contract status, and client type from historical applications
• Correlation heatmap for previous application dataset
• Scatter plot confirming linear relationship between AMT_APPLICATION and AMT_GOODS_PRICE

---

🛠️ Tech Stack

Tool	Purpose
Python 3	Core analysis language
Pandas	Data manipulation and cleaning
NumPy	Numerical operations
Matplotlib	Visualization
Seaborn	Statistical plots and heatmaps
Jupyter Notebook	Interactive analysis environment

---

🚀 Getting Started

# Clone the repository
git clone https://github.com/akshitnlabs/<repo-name>.git
cd <repo-name>

# Install dependencies
pip install pandas numpy matplotlib seaborn jupyter

# Launch the notebook
jupyter notebook EDA_Case_Study_Final.ipynb

Note: Place application_data.csv and previous_application.csv in the same directory as the notebook before running.

---

👥 Authors

• Akshit Nair — GitHub
• Anand Mohan

---

📌 Key Takeaways

• Clients with lower education levels and working income type were more represented among defaulters
• Loan period and annuity amount showed notable distributional differences between risk segments
• A strong linear relationship exists between AMT_GOODS_PRICE and AMT_APPLICATION in previous applications, suggesting consistent pricing behavior
• Age is a meaningful risk differentiator — younger borrowers showed higher default rates

---

This project was completed as part of a data analytics case study to develop practical EDA skills on real-world financial data.