3.2. Core Machine Learning Concepts

💡 First Principle: Features are inputs (what you know), labels are outputs (what you're predicting). This distinction is foundational—every ML problem reduces to learning a function that maps features to labels. Get this wrong and nothing else makes sense.

What breaks without this clarity: Exam questions often describe a scenario with multiple data points and ask "how many features and labels?" If you confuse inputs with outputs, you'll count wrong. Features are the descriptive attributes you feed IN; the label is the prediction that comes OUT. One common trap: "predicting diabetes risk from age, weight, and blood pressure" has THREE features (age, weight, BP) and ONE label (risk).

Think of features and labels like exam questions and answers. Features are the questions—the information provided. Labels are the answers—what you're trying to determine. In supervised learning, you study with the answer key (labeled data) so you can answer new questions (predict labels for new features). The model learns the pattern connecting questions to answers.

Now that you understand ML techniques, let's examine the fundamental concepts that apply to all machine learning: features, labels, and dataset management.