End-to-End Churn Model

What you'll learn

• Performing Train/Test/Validation splits correctly
• Training a baseline classifier vs. a more advanced ensemble model
• Evaluating models using Precision, Recall, and F1-score
• Saving/Exporting a model and serving predictions over an API

Architecture

Tabular Data (Kaggle)
  → Preprocessing Pipeline (Imputing, Scaling, Encoding)
  → Model Training (scikit-learn RandomForest)
  → Export (Joblib)
  → FastAPI Model Server

Steps

1. Find a customer churn dataset on Kaggle (e.g., Telco Customer Churn).
2. Clean the data and split it into training and testing sets.
3. Build a scikit-learn Pipeline that handles scaling numerical features and one-hot encoding categorical features.
4. Train a baseline Logistic Regression model, then train a Random Forest classifier.
5. Evaluate both models using a confusion matrix, precision, recall, and F1-score. Pick the best one.
6. Export the fitted model and pipeline using joblib.
7. Write a minimal FastAPI app that loads the joblib file on startup and exposes a /predict endpoint taking user features and returning a churn probability.