About This Project
Overview
A PyTorch neural network trained on historical customer data to predict churn likelihood, giving businesses a ranked view of which customers are at risk before they leave.
Motivation
Churn prediction is a classic applied ML problem, and I wanted a from-scratch PyTorch implementation to really understand what a production churn model has to account for.
Lessons Learned
Feature engineering — recency, frequency, and engagement trends — moved the needle far more than model architecture; a simple network with good features outperformed a more complex one with raw features.