Credit Risk Classification Pipeline

A modular machine learning pipeline designed to predict loan default risk using structured tabular data. The system reflects industry-standard approaches used in credit scoring and financial risk assessment.

Project Overview

• Objective: Binary classification of loan default risk
• Dataset: Kaggle Credit Risk Dataset
• Target Variable: loan_status
• Features: Borrower demographics, loan characteristics, credit history

Pipeline Structure

Modularized for clarity and extensibility:

• load_data.py: Downloads and loads the dataset
• preprocess.py: Handles missing values, encodes features, removes outliers
• train.py: Trains models using Scikit-learn Pipelines with scaling
• evaluate.py: Generates performance metrics and visualizations
• inference.py: Loads saved models for prediction
• scripts/run_pipeline.py: Executes the full end-to-end workflow

Data Preprocessing

Key data cleaning and feature engineering steps include:

• Missing Values:

  • Imputed missing employment length with 0 and flagged with a separate binary indicator
  • Dropped rows missing loan interest rates due to contextual inconsistency

• Outlier Removal:

  • Removed unrealistic ages (>122)
  • Excluded records indicating employment before age 13

• Feature Encoding:

  • One-hot encoded nominal features (person_home_ownership, loan_intent, cb_person_default_on_file)
  • Ordinally encoded loan_grade (A–G → 1–7)

• Stratified Sampling:

  • Ensured class distribution was preserved in training and testing sets

Models Implemented

• Logistic Regression
• Random Forest
• Gradient Boosting

Evaluation

• Stratified train/test split to preserve class balance
• Metrics reported: Accuracy, Precision, Recall, F1 Score, ROC AUC
• Visual outputs: Confusion matrix, ROC curve, Precision-Recall curve

Outputs

• Trained models saved to /models/
• Evaluation plots saved to /outputs/
  • 
  • 
  • 

Key Highlights

• Fully modular and reproducible implementation using Scikit-learn Pipelines
• Incorporates best practices for preprocessing, tuning, and model evaluation
• Suited for credit risk prediction and similar tabular classification tasks

How to Run

# Run the complete training and evaluation pipeline
python -m scripts.run_pipeline