Kaggle Learn: Intro to Machine Learning

Roughly three hours, six lessons, and a trained model at the end of the first one. Kaggle's introduction is built around the observation that most people bounce off machine learning during the setup rather than the maths, so it removes setup entirely — everything runs in hosted notebooks with the data already loaded.

The lessons run in a tight sequence: How Models Work, Basic Data Exploration, Your First Machine Learning Model, Model Validation, Underfitting and Overfitting, and Random Forests. Every one is a short reading followed by an exercise notebook you actually complete. You'll need a Kaggle account to save work and to receive the completion certificate, which is free.

What it gets right

The order. Validation comes before you learn a second algorithm, which is exactly backwards from how most courses do it and exactly right. A beginner who understands why a train-test split exists is more useful than one who can name six models, and the Underfitting and Overfitting lesson lands harder for coming immediately after.

It's also pandas and scikit-learn from the first cell, which is the stack you'd actually use.

How shallow is it, really

Very. That's the design, but be clear-eyed about it. There is no maths — no gradient descent, no loss functions, no derivation of anything. Decision trees and random forests are presented as tools you call, not mechanisms you understand. Feature engineering barely appears. Nothing about classification metrics, imbalanced data, cross-validation strategies, or what to do when your model works in a notebook and fails on real inputs.

Three hours also means the exercises are mostly fill-in-the-blank. You'll write real code, but rarely more than a line or two at a time, so the feeling of competence outruns the actual skill by some margin. The certificate reflects completion, nothing more — it is not accredited and carries no weight in hiring.

Who it's for — and who should skip it

Ideal if you write Python, keep meaning to try machine learning, and want to find out in one evening whether you like it. Also a reasonable pre-read before a heavier course, since the vocabulary you pick up here makes fast.ai or a university course less alienating.

Skip it if you want to understand what the algorithms are doing — go to StatQuest or a proper course. Skip it too if you're already comfortable with scikit-learn, since there's nothing here you don't know.

Our take

The best three-hour on-ramp to machine learning we know, and a bad place to stop. The risk is precisely that it feels complete: it hands you a working random forest without teaching you enough to know when to distrust it. Take it as a taster and immediately follow it with something that shows the maths.