Learning objectives
- Understand classification as a supervised task
- Recognize binary, multiclass, and multilabel problems
- Identify practical examples of classification
Introduction
Classification is one of the most common AI tasks. In a classification problem, the model predicts a category rather than a number. This category might be simple, such as yes or no, or more complex, such as one of many product types, medical conditions, or document topics.
Classification is widely used because many decisions are categorical. Should a transaction be flagged or not? Which department should handle this ticket? Is this review positive, negative, or neutral?
For beginners, the main skill is recognizing when a problem naturally fits classification and understanding the differences among common forms of classification.
Binary, multiclass, and multilabel
Binary classification involves two categories, such as fraud versus legitimate or pass versus fail. Multiclass classification involves more than two categories, such as predicting whether an image shows a cat, dog, bird, or horse.
Multilabel classification is different because one item can belong to several categories at the same time. A news article might be tagged as politics, economy, and international affairs simultaneously.
Inputs, outputs, and confidence
Classification models often return both a predicted class and a confidence score or probability. This helps humans decide whether to trust the output or send uncertain cases for review.
In sensitive applications, using thresholds and escalation paths is often better than blindly acting on every automated prediction.
Why class balance matters
Some classification tasks involve highly imbalanced classes. Fraud may be rare compared with legitimate transactions, and disease positives may be much fewer than negatives. In such cases, overall accuracy can be misleading.
Practitioners must pay attention to class distribution, evaluation metrics, and the costs of different errors.
Examples
Spam detection
A model predicts whether each incoming email belongs to the spam or non-spam category.
Ticket routing
A support center classifies messages as billing, shipping, technical, or account-related so they reach the correct team quickly.
Image recognition
A classroom project classifies photos of animals into several species categories.
Exercises
- Give one example each of binary, multiclass, and multilabel classification.
- Why can accuracy be misleading in rare-event classification?
- Describe a classification problem from your own field of interest.
- Why might a confidence score be useful alongside the predicted class?
- Turn three everyday decisions into possible classification tasks.
Key takeaway
Classification helps AI turn messy inputs into useful categories, making it central to many business, educational, and public-service applications.