Supervised vs Unsupervised Learning

Supervised learning means you have labels. You show the model houses with their prices, and it learns to predict price for new houses. Unsupervised means you have no labels, and the model finds hidden groups, like clustering customers by buying habits.

Supervised splits into classification (predict a category, like spam or not spam) and regression (predict a number, like price). Unsupervised is mostly clustering and dimensionality reduction. Choosing the wrong type is a common beginner error.

If you have a target column, you are supervised. If you only have features and want to find patterns, you are unsupervised. Everything from evaluation to deployment changes based on this choice.

How to tell which you need

Ask: do I have a column I want to predict? If yes, supervised. If you just want to group similar rows, unsupervised. Mix them and your metrics will be meaningless.

# supervised: X + y -> model
# unsupervised: X -> clusters
# y is None for unsupervised

Labels are expensive

Supervised needs labeled data, which often requires humans to annotate. Unsupervised can run on raw data but its results are harder to evaluate. This tradeoff decides project cost.

# classification: y = ["spam","ham"]
# regression: y = [200000, 350000]
# clustering: no y

Syntax

# Supervised
from sklearn.ensemble import RandomForestClassifier
model = RandomForestClassifier()
model.fit(X_train, y_train)  # y exists

# Unsupervised
from sklearn.cluster import KMeans
kmeans = KMeans(n_clusters=3)
kmeans.fit(X)  # no y

# Don't: kmeans.fit(X); y_fake = kmeans.labels_; clf.fit(X, y_fake)  # circular