🏦 Loan Default Prediction: Probability of Default (PD)

Financial institutions face significant risks due to loan defaults. Accurately predicting the Probability of Default (PD) is critical for effective risk management and strategic planning.

🎯 In this project, we develop a machine learning model to estimate the likelihood that a loan will default, based on historical loan data.

---

📊 Problem Statement

Build a predictive model to estimate the probability of default for loan applicants using historical loan application data.

---

🧭 My Approach

📦 1. Package Imports

• Imported all necessary libraries including pandas, numpy, matplotlib, seaborn, sklearn, shap, etc.
• Managed long list of packages to support EDA, preprocessing, modeling, and interpretation.

---

🧮 2. Data Loading & Cleaning

• Fixed index column issues in the dataset.
• Dropped repeated columns that appeared due to merging or saving errors.

---

🔎 3. Exploratory Data Analysis (EDA)

✅ Cleaned Key Columns:

• remaining_term: Standardized inconsistent values.
• job: Unified titles and standardized missing entries.
• location: Cleaned and grouped underrepresented categories to avoid data sparsity.
• currency and country: Dropped due to low variance and redundancy.
• gender: Added an explicit NaN category to "Other" for clarity.

---

🏗️ 4. Feature Engineering

• Created new features based on correlation analysis with the target (default).
• Engineered interaction features where relevant.
• Ensured no data leakage during feature creation.

---

🔧 5. Preprocessing

• 🧠 Gender Imputation: Used a Random Forest classifier to intelligently impute missing values.
• 🧹 Simple Imputer: Applied for other missing categorical/numerical values.
• Scaled numerical features and one-hot encoded categoricals.

---

🧪 6. Model Selection

• Evaluated multiple models using cross-validation:
  • Logistic Regression
  • Random Forest

---

🧬 7. Feature Selection

• Tried Recursive Feature Elimination (RFE) but results were unsatisfactory.
• Opted for manual selection based on domain knowledge and model performance.

---

🛠️ 8. Hyperparameter Tuning

• Used RandomizedSearchCV for fast and efficient parameter optimization.

---

🧠 9. Interpretation

• Applied SHAP (SHapley Additive exPlanations) to interpret model predictions.
• Identified key features driving loan default risk:
  • Loan amount
  • Remaining term
  • Employment status
  • Income vs. loan ratio

---

🔒 Note: This project is for educational purposes and should not be used as financial advice or in production without further validation.