🧪 Feature Engineering Playground

A Microservices-based ML Application for Churn Prediction

---

📌 Project Overview

This project demonstrates the Data-Centric AI philosophy.
Instead of repeatedly changing machine learning algorithms, the model (XGBoost) is kept fixed while performance is improved through systematic feature engineering and cost-sensitive learning.

The system follows a Microservices-lite architecture:

• Backend: FastAPI (ML inference & logic)
• Frontend: Streamlit (Interactive dashboard)
• Experiment Tracking: MLflow (Metrics, parameters, artifacts)

---

📸 Experiment Results & Insights

1️⃣ Feature Strategy Analysis (Parallel Coordinates)

Visualizing how feature count and class-weight tuning impact the F1 score.

Key Insight — Curse of Dimensionality

The highest F1 score (~0.635) was achieved using a balanced feature set.
Enabling all engineered features (Polynomial + Interaction + Binning) caused a slight performance drop (~0.634), showing that adding noisy or redundant features can hurt generalization even in robust models like XGBoost.

---

2️⃣ Feature Importance & Model Performance

Engineered features consistently outperform raw inputs.

Performance Summary

• Baseline F1: ~0.58 (Standard XGBoost)
• Final F1: ~0.635 (Feature Engineering + Cost-Sensitive Learning)
• ROC-AUC: ~0.84

---

3️⃣ Live Inference Pipeline

Trained models are persisted and served for real-time predictions.

---

🏗️ Project Architecture

├── data/               # Raw and processed datasets
├── src/
│   ├── backend/        # FastAPI server (ML inference)
│   ├── frontend/       # Streamlit UI
│   └── core/           # Feature engineering and training logic
├── assets/             # Screenshots and diagrams
├── notebooks/          # EDA and experimentation
└── requirements.txt    # Python dependencies

---

🚀 Getting Started (Local Setup)

Follow the steps below to run the complete system locally, including the backend API, frontend dashboard, and experiment tracking.

1️⃣ Environment Setup

# Clone the repository
git clone https://github.com/DS-ProjectRepo/feature-playground.git
cd feature-playground

# Create virtual environment
python -m venv venv
source venv/bin/activate        # On Windows: venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt

---

2️⃣ Run the Application

Run the backend and frontend in separate terminals.

Terminal 1 — Backend (FastAPI)

uvicorn src.backend.main:app --reload

Terminal 2 — Frontend (Streamlit)

streamlit run src/frontend/app.py

---

3️⃣ Experiment Tracking (MLflow)

To view experiment history and metrics:

mlflow ui

Open http://localhost:5000 in your browser.

---

🧠 Engineering Concepts Applied

1. Interaction Features Multiplicative combinations (e.g., Tenure × MonthlyCharges) to capture non-linear relationships.

2. Polynomial Expansion Adds curvature to numeric features to improve expressiveness.

3. Domain-Driven Binning Converts continuous variables (e.g., tenure) into meaningful cohorts.

4. Cost-Sensitive Learning Adjusts scale_pos_weight in XGBoost to handle class imbalance (~1:3 churn ratio).

5. Microservices Pattern Decouples inference logic from the user interface using REST APIs.

---

📌 Key Takeaway

Feature quality matters more than model complexity. Thoughtful feature engineering and disciplined evaluation consistently outperform aggressive model tuning.

---

📄 License

This project is released for educational and research purposes.