What is supervised learning & unsupervised learning?

Imagine for every problem that you have, you have someone (like a teacher) who tells all the right answers. That is supervised learning. On the other hands, unsupervised learning means you have to answer on your own without any guidance. This is the simplified way to explain. So, what about more formal definitions?

Supervised learning is a method using labeled data. It finds a prediction function given features and associated labels. The prediction function is used to predict the label of features never seen.

Unsupervised learning is a method using unlabeled data. It finds a prediction function given features without associated labels. The prediction function is used to cluster or group the label of features never seen.

Here are the key main differences between supervised learning and unsupervised learning:

Supervised learning	Unsupervised learning
Input data are labeled.	Input data are NOT labeled.
Uses training data sets	Uses input data sets
For classification, regression, and ranking purposes.	For analysis purposes.

With supervised learning, you can do three main things, which are classification, regression, and ranking.

• Classification: you have a finite set of labels. E.g.: face recognition, finding low-risk & high-risk loans, spam detection, medical diagnosis.
• Ranking: labels are a ranking number. E.g.: movie ranking in Netflix, Google’s web pages ranking, flight search.
• Regression: labels are real-valued numbers. E.g.: predict stock values, find weather’s temporal trends over time, predict price of a used car.

With unsupervised learning, you can think of the following applications: customer segmentation (find who are most likely to purchase again), image compression, and data dimensionality reduction.