Supervised, Unsupervised and Reinforcement Learning

Labels, structure, rewards

Supervised learning uses labelled examples: regression predicts a number (house price, delivery time) and classification predicts a category (spam or not, which flower). Unsupervised learning has no labels and finds structure: clustering groups similar items, dimensionality reduction compresses features, anomaly detection finds unusual items. Reinforcement learning learns actions by trial and error from rewards (games, robotics, some recommendation and control systems). Most business ML is supervised, so most of this course focuses there.

Problem types with examples

Match the question to the family.

question                                        family           output
how much will this house sell for?             regression        a number
is this transaction fraudulent?                classification    a class (yes/no)
which customers behave alike?                  clustering        group ids
which sensor readings are unusual?             anomaly detection a flag / score
how should a robot move to reach the goal?     reinforcement     actions

Phrase the question precisely

Write the exact prediction you need, its unit and when it must be available; vague goals lead to the wrong problem type.

Quick check: Predicting tomorrow's sales in rupees is which kind of problem?

• Regression
• Classification
• Clustering
• Reinforcement learning