Types of Machine Learning

Machine learning approaches are commonly categorized by how the model receives feedback or learns from data.

Supervised Learning

Supervised learning uses labeled examples. Each training example contains input features and a known target value or class.

For example, a house-price dataset might contain:

The model learns relationships between the features and the target price.

Supervised learning includes two common problem types:

Regression: Predicts a numerical value, such as a house price or delivery time.

Classification: Predicts a category, such as spam versus legitimate email or fraudulent versus legitimate transactions.

Common supervised learning algorithms include linear regression, logistic regression, decision trees, random forests, and support vector machines.

Unsupervised Learning

Unsupervised learning works with data that does not have predefined target labels. The model attempts to identify structure, patterns, or relationships in the data.

For example, an online retailer might group customers based on purchasing behavior without assigning customer categories beforehand.

Common tasks include:

• Clustering: Grouping similar observations.
• Dimensionality reduction: Representing data using fewer dimensions while preserving useful information.
• Anomaly detection: Identifying observations that differ significantly from typical patterns.

Common algorithms include K-means clustering, hierarchical clustering, and principal component analysis (PCA).

Unsupervised learning can reveal useful patterns, but the resulting groups or patterns still need interpretation.

Reinforcement Learning

Reinforcement learning trains an agent to make decisions by interacting with an environment and receiving rewards or penalties.

The agent observes a situation, chooses an action, and receives feedback. Over time, it learns a policy intended to maximize cumulative reward.

Examples include:

• Training an agent to play a game.
• Optimizing certain robotic movements.
• Learning strategies in simulated environments.
• Managing resource allocation under defined objectives.

Unlike standard supervised learning, reinforcement learning does not necessarily require a labeled correct action for every situation.

Comparing the Three Approaches

Choosing the Right Approach

Suppose you are developing different applications:

• Predicting employee salaries from historical examples is a supervised regression problem.
• Grouping customers based on their purchasing behavior is an unsupervised learning problem.
• Training a simulated robot to navigate an environment using rewards is a reinforcement learning problem.

The right approach depends on the objective, the available data, and the type of feedback the system can receive.