Machine Learning Tutorial 0/98 lessons ~6 min read Lesson 3
Types of Machine Learning
ML problems fall into a few recurring families.
Course progress0%
Focus
5 guided sections
Practice signal
Examples included
Career prep
Foundation builder
Introduction
ML problems fall into a few recurring families. Picking the right family determines which algorithms, metrics, and unlabeled data you can use.
Understanding the topic
Major paradigms:
- Supervised — labeled rows; predict class or numeric target.
- Unsupervised — no labels; discover structure such as clusters or low-dimensional views.
- Reinforcement — sequential actions with scalar rewards; learn a policy.
- Self-supervised — labels manufactured from the data itself (common in modern representation learning).
- Semi-supervised — small labeled set plus large unlabeled pool.
| Paradigm | Typical data | Objective | Example use |
|---|---|---|---|
| Supervised | Labeled table | Predict Y from X | Churn, pricing |
| Unsupervised | Unlabeled table | Structure / compression | Segments, PCA |
| Reinforcement | Trajectories + reward | Maximize return | Robotics, games |
| Semi-supervised | Few labels, many unlabeled | Label efficiency | Medical scans |
Execution workflow
1Choosing a paradigm
1 / 4Labels available?
Yes → start supervised (classification or regression).
Best practices
- Most business tabular tasks begin supervised.
- Run unsupervised EDA even when labels exist — it surfaces leakage and bad features.
Summary
Match the learning paradigm to the data you can afford to collect and the decision you must automate.
Ready to mark this lesson complete?Track your journey across the entire course.