What is supervised learning?
Supervised learning trains a model on labeled examples — inputs paired with the right answer — so it can predict the answer for new inputs. Regression predicts a number, classification predicts a category. It is the workhorse of practical ML.
Why it matters
Most ML problems that deliver business value — predicting churn, classifying images, scoring leads — are supervised. Understanding how a model learns from labels, and the core algorithms, gives you the foundation that deep learning later builds on. It is where every ML practitioner starts.
What to learn
- Features, labels, and the train/predict cycle
- Regression: linear and beyond
- Classification: logistic regression, trees, k-NN
- The decision boundary intuition
- Training, generalization, and the bias-variance trade-off
- Overfitting versus underfitting
- The role of the loss function
Common pitfall
Judging a model by its accuracy on the data it trained on. A model can memorize the training set and look perfect while failing on anything new — that is overfitting. Always measure performance on held-out data the model has never seen, because training-set accuracy tells you almost nothing about real performance.
Resources
Primary (free):
- Google — Machine Learning Crash Course · docs
- scikit-learn — Supervised learning · docs
- StatQuest — ML playlist · video
Practice
Train a regression model to predict a numeric target and a classifier to predict a category, each on a small labeled dataset. Hold out a test set and report performance on it, not on training data. Done when you can explain whether each model is overfitting based on the train-versus-test gap.
Outcomes
- Distinguish regression from classification tasks.
- Train a model on labeled data and predict on new inputs.
- Explain overfitting and the bias-variance trade-off.
- Always evaluate on held-out data, not the training set.