🧩
Problem Framing Lab
From: Framing Problems•View in lesson →
💡
Before you can solve a problem with ML, you need to frame it correctly. Is it classification or regression? What are your features? What are you predicting? This lab guides you through the crucial first step of any ML project.
Customer Churn Prediction
E-commerce company wants to reduce customer loss
Available Data:
10,000 customers, features: purchase history, support tickets, last login, account age
Your Task:
Choose how to formulate this as an ML problem. Different formulations lead to different models and outcomes.
How should you frame this problem?
Key Insights About Problem Framing
- 1.Same data, different questions: You can formulate many different ML problems from the same dataset
- 2.Business goal drives formulation: Start with “What decision will this enable?” not “What can I predict?”
- 3.Proxy metrics are dangerous: Predicting something correlated isn't the same as predicting the right thing
- 4.Simpler often better: Binary classification often beats multi-class for initial deployment
- 5.Most ML failures: Happen in problem formulation, not model choice
Experiment as much as you like; there is no progress to lose.