Supervised learning
Supervised learning is the most widely used category of machine learning. The defining characteristic is that every training example comes with a label —...
Learning from Labelled Examples
Supervised learning is the most widely used category of machine learning. The defining characteristic is that every training example comes with a label — the correct answer the model should learn to predict. The algorithm adjusts itself to match its predictions to those labels as closely as possible, then applies what it learned to unlabelled new examples.
Supervised learning divides into two main problem types. Classification assigns examples to discrete categories. Email spam detection is classification — each email is either spam or not spam. Image recognition is classification — each image belongs to a class like cat, dog, or car. Regression predicts a continuous numerical value. Estimating the selling price of a house is regression — the output is a number on a continuous scale.
The training process is guided by a loss function that quantifies how far predictions are from true labels. For classification this might be cross-entropy loss; for regression it is often mean squared error. Optimization algorithms like gradient descent iteratively move the model's parameters in the direction that reduces the loss.
Supervised learning requires labelled data, and labelling can be expensive. Medical imaging datasets, for instance, require expert radiologists to annotate thousands of scans. The cost of labelling is one practical reason practitioners also study semi-supervised and unsupervised approaches.
A concrete example: a bank trains a supervised model to approve or reject loan applications. Historical records contain each applicant's features (income, credit score, employment history) and whether their loan was eventually repaid or defaulted. The model learns to predict repayment probability for new applicants.
Exam tip: know that supervised learning requires labelled data, and be ready to distinguish classification (discrete output) from regression (continuous output) with one example of each.