Credit Risk Scoring System (ML + XAI + MLOps)

Overview

This project implements an end-to-end machine learning system for credit risk scoring using the German Credit dataset. It covers data preprocessing, model training, evaluation, explainability, experiment tracking, and deployment via a REST API.

The system compares multiple machine learning models and provides interpretability using SHAP and LIME.

---

Problem Definition

The goal is to predict credit risk:

• 0 → Bad credit risk
• 1 → Good credit risk

---

Machine Learning Models

The following models are implemented:

• Logistic Regression (baseline model)
• Random Forest
• XGBoost

---

Evaluation Metrics

Model performance is evaluated using:

• Accuracy
• ROC-AUC
• Confusion Matrix
• ROC Curve

---

Explainability (XAI)

Model interpretability is provided using:

• SHAP (global and local explanations)
• LIME (local explanation for individual predictions)

---

MLOps Components

• MLflow for experiment tracking:
  • parameters
  • metrics
  • artifacts
  • trained models
• Reproducible training pipeline
• Structured modular codebase

---

API Service

A FastAPI-based inference service is implemented.

Endpoint

POST /predict

Request Body

{
  "status_checking_account": "A11",
  "duration_months": 12,
  "credit_history": "A34",
  "credit_amount": 2500,
  "age": 35
}

Response

{
  "risk_score": 0.73,
  "decision": "reject"
}

Tech Stack

• Python
• Scikit-learn
• XGBoost
• MLflow
• SHAP
• LIME
• FastAPI
• Pandas
• NumPy

---

How to Run the Project

The system has three main components that must be run separately:

1. Run ML Training Pipeline

This runs model training, evaluation, and MLflow logging.

python main.py

Run MLflow Tracking Server

mlflow ui --host 127.0.0.1 --port 5000

Then open in browser:

http://127.0.0.1:5000

Run FastAPI Service

uvicorn api_:app --reload

Default endpoint:

http://127.0.0.1:8000

Swagger UI:

http://127.0.0.1:8000/docs