Classification

Definition

A machine learning task where the model assigns input data to predefined categories. Spam detection (spam or not spam), image recognition (cat or dog), and sentiment analysis (positive or negative) are all classification problems. One of the most common and practical applications of ML.

How It Works

Classification is one of the most fundamental tasks in machine learning: take some input and sort it into predefined categories. Email spam filters classify messages as spam or not spam. Image recognition classifies photos as containing cats, dogs, or neither. Sentiment analysis classifies text as positive, negative, or neutral.

There are two main flavors. Binary classification has two categories (yes/no, spam/not-spam). Multi-class classification has many categories (classifying a news article as sports, politics, tech, or entertainment). The model learns the boundaries between categories from labeled training data — examples where humans have already assigned the correct category.

Before large language models, classification required training a dedicated model for each specific task. Now, LLMs can classify text with just a prompt: "Is this review positive or negative?" This zero-shot classification ability has democratized the technology — you don't need a data science team to build a classifier anymore. For high-stakes applications though, dedicated fine-tuned classifiers still tend to be more reliable and faster.