Commerce Decisions
Churn Prediction API
A from-scratch logistic regression returning churn probability, risk tier and top contributing factors.
The Problem
Retention teams need to know who is at risk and why, so the action can match the cause.
How It Works
- Train offline (Java trainer) on seven standardised customer features and commit the weights.
- Serve p = sigmoid(bias + Σ w·z).
- Return the risk tier (<0.30 low, <0.60 medium) and the three largest signed contributions.
Technically Interesting
The model is interpretable by construction: each factor’s contribution is its weight times its standardised value.
Why It Matters
A small, explainable model is often the right answer in CRM, and it is the one genuine machine-learning component in this lab.
Scope & honesty. Trained on synthetic data; a demonstration of the technique, not a validated production model.