Until now, we have learned how to:

Define business problem
        ↓
Identify features and label
        ↓
Check data quality and leakage
        ↓
Create training, validation and test sets

But not every ML problem contains a label, and not every model simply predicts true or false.

Consider three requirements:

1. Predict whether a policy will renew.
2. Discover natural customer groups.
3. Learn which action produces the best long-term result.

These requirements use different learning approaches:

Requirement Learning approach
Predict a known outcome Supervised learning
Discover hidden groups or structure Unsupervised learning
Learn through actions and rewards Reinforcement learning

Today’s core principle:

First identify the type of learning problem. Algorithm selection comes afterward.

PreviousNext