v0.1.0 — End-to-End Credit Risk Engine (Experimentation Release)

Overview

Initial end-to-end release of the Credit Portfolio Risk Engine.

This version demonstrates a complete machine learning workflow, starting from raw credit portfolio data and ending with a deployable prediction service and interactive dashboard.

The project combines feature engineering, model experimentation, model serving, experiment tracking, containerization and deployment concepts into a single reproducible data product.

---

Features

Data Engineering

• Raw credit portfolio dataset ingestion
• Bureau A1 feature integration
• Bureau A2 feature integration
• Feature aggregation and dataset preparation
• Missing value handling

Machine Learning

• Baseline model experimentation
• Hyperparameter optimization with Optuna
• XGBoost model training
• Feature importance analysis
• Validation AUC and Gini evaluation

MLOps

• MLflow experiment tracking
• Model artifact management
• Training metadata logging
• Reproducible experiment runs

API Layer

• FastAPI REST API
• Health check endpoint
• Prediction endpoint
• JSON request/response schema

User Interface

• Streamlit dashboard
• Interactive risk scoring
• Probability of Default visualization
• Risk band classification
• Prediction summary reporting

Deployment

• Dockerized API service
• Dockerized Streamlit application
• Docker Compose orchestration

---

Model Performance

Metric	Value
Validation AUC	0.824
Validation Gini	0.648
Features Used	258
Model	XGBoost

---

Architecture

Raw Data
    ↓
Feature Engineering
    ↓
Training Dataset
    ↓
XGBoost Training
    ↓
MLflow Tracking
    ↓
FastAPI REST Service
    ↓
Streamlit Dashboard
    ↓
Docker Deployment

---

Technology Stack

• Python
• Pandas
• NumPy
• Scikit-Learn
• XGBoost
• Optuna
• MLflow
• FastAPI
• Streamlit
• Docker

---

Notes

This release focuses on experimentation, learning, and end-to-end integration.

Future releases may include:

• SHAP explainability
• Feature store integration
• Model monitoring
• Data drift detection
• CI/CD pipelines
• Automated retraining workflows
• Cloud deployment

---

Highlights

• End-to-end machine learning workflow
• Real-world credit risk dataset
• 1.5M+ customer records processed
• 258 engineered features
• MLflow experiment tracking
• FastAPI model serving
• Streamlit business dashboard
• Docker deployment ready