Stop 06 · Machine Learning FundamentalsTopics · 1/13

Supervised vs Unsupervised vs Reinforcement Learning

In short

Every machine learning problem starts with one question: what kind of learning is this? The answer decides which algorithms, data and metrics you'll use.

Supervised learning learns from labeled examples (input → known output), unsupervised learning finds structure in unlabeled data, and reinforcement learning learns through trial, error and reward.

The mascot holds a picture flashcard of a mango next to a real mango, with a small bicycle leaning nearby.

Think of it like this

Three ways a child learns

A parent shows flashcards: 'this is a mango'. That's supervised, learning from labelled answers. The child sorts a box of buttons into piles by colour with no names given: unsupervised. Learning to ride a cycle by wobbling, falling and trying again: reinforcement.

Watch it happen

The three families

Machine learning

Supervised

Unsupervised

Reinforcement

Classification

Regression

Clustering

Dim. reduction

Game agents

1/4Three families, split by what feedback the model gets while learning.

The big ideas

What you need to know

Labels make it supervised

If your data has a target column you want to predict, it's supervised learning.

No labels, find structure

Unsupervised methods group or compress data. A human still decides what the groups mean.

Rewards drive RL

Reinforcement learning learns from trial and error. Common in games and robotics, rarer in everyday business work.

Most jobs are supervised

Churn, fraud, demand and pricing models are mostly supervised. Start your practice there.

See it in code

Walk through the code

python

1from sklearn.datasets import load_iris2from sklearn.linear_model import LogisticRegression3from sklearn.cluster import KMeans4 5X, y = load_iris(return_X_y=True)6 7clf = LogisticRegression(max_iter=1000).fit(X, y)   # supervised: uses labels y8print(clf.predict(X[:3]))9 10km = KMeans(n_clusters=3, n_init=10, random_state=0).fit(X)   # unsupervised: no y11print(km.labels_[:3])

1/3X holds flower measurements, y holds the species labels.

Same flower measurements, two kinds of learning. One learns from the species labels, the other finds groups on its own.

Save yourself the pain

Common mistakes

Oops

Thinking clustering will tell you which customers will churn.

Do this instead

Churn is a labelled outcome, so it's a supervised problem. Clustering only groups similar customers.

Oops

Reaching for reinforcement learning for a normal prediction task.

Do this instead

If you have historical examples with answers, supervised learning is simpler and usually better.

Your mission

Classify ten problems

~15 min

Quick check

You have past loan applications marked 'repaid' or 'defaulted'. Which type of learning fits?

Go deeper

1Classify 5 real-world ML applications you know into these 3 categories.
2Learn why most business data science work is supervised learning.
3Understand reinforcement learning conceptually even if you won't use it day-to-day.