3.1. Machine Learning Techniques

💡 First Principle: Machine learning techniques divide into three categories based on one question: "Do you have labels (correct answers) in your training data?" If yes, you're doing supervised learning (regression or classification). If no, you're doing unsupervised learning (clustering). This single question eliminates half your options instantly.

What breaks without this understanding: The exam presents scenarios like "predict house prices" or "group customers by behavior" and expects you to choose the right technique. Without the labels/no-labels framework, regression, classification, and clustering blur together. With it, the choice becomes mechanical: labels + numeric output = regression, labels + categories = classification, no labels = clustering.

Think of it like cooking. Supervised learning is following a recipe (labels = instructions telling you what the dish should become). Unsupervised learning is being handed random ingredients and asked to invent something—you discover patterns yourself. The exam constantly tests whether you can identify which situation a scenario describes.

Building on the supervised vs unsupervised distinction from Section 1.3, let's examine specific ML techniques. The key is matching the prediction type to the correct algorithm category.

Classification vs Regression Decision: