Machine Learning Crash Course
Build, test and ship honest machine learning models with scikit-learn.
- 4.9 (1,090 ratings)
- 9,600 learners
- Intermediate
- 7h 25m total
- Updated August 2026
What you’ll learn
- Frame a business question as a supervised learning problem
- Train and compare models with fit, predict and score
- Choose metrics that match the cost of each kind of mistake
- Build leak-free pipelines with ColumnTransformer and cross-validation
- Tune hyperparameters and engineer features that move the score
- Check a model for bias, monitor drift and ship it behind an API
Who it’s for
Most machine learning tutorials stop at a high accuracy score. This course starts where real projects start: a business question, a messy table and a baseline that any model has to beat. You predict which customers of Cartwheel, an online store, will stop ordering, and you learn regression, classification, trees, ensembles and neural networks by building them in scikit-learn.
Every idea comes with a picture, a few lines of code you run in the browser and a check on what the numbers really mean. You learn the habits that separate a demo from a model people can trust: cross-validation, pipelines, leakage hunting, threshold choices, fairness checks and monitoring after launch. The maths is shown with code and diagrams, not pages of notation.
- Who it’s for
- Analysts and developers who know some Python and pandas and want to build machine learning models they can defend.
- Before you start
- Basic Python and pandas: loading a CSV, selecting columns and filtering rows. No maths beyond school algebra.
Hands-on practice
- Guided exercises: spot the bug, fill the command, order the steps.
- Run real Python and pandas in your browser — no install.
Course content
5 sections · 23 lessons · 7h 25m total
- Learning from Examples14 min
- The Workflow and the Train/Test Split18 min
- Models in scikit-learn: fit, predict, score18 min
- Baselines: The Model to Beat17 min
- Section checkpoint5 questions
- Linear Regression and Its Metrics22 min
- How Models Learn: Loss and Gradient Descent22 min
- Polynomial Features and Overfitting20 min
- Regularisation with Ridge and Lasso20 min
- Section checkpoint5 questions
- Logistic Regression: Predicting Probabilities20 min
- Decision Trees18 min
- Random Forests and Gradient Boosting20 min
- Classification Metrics Beyond Accuracy22 min
- Thresholds and Class Imbalance20 min
- Section checkpoint5 questions
- Cross-Validation18 min
- Pipelines and ColumnTransformer22 min
- Hyperparameter Search20 min
- Data Leakage18 min
- Feature Engineering for Churn20 min
- Section checkpoint5 questions
- Clustering and PCA21 min
- Neural Networks and Deep Learning20 min
- Where LLMs and Embeddings Fit17 min
- Responsible ML: Bias, Fairness and Drift20 min
- Shipping a Model18 min
- Section checkpoint5 questions
Your instructor
Liam Patel
Lead Data Scientist · Datawheel
- 4.7 instructor rating
- 32,900 learners
- 3 courses
Liam started as a business analyst who got tired of copying numbers between spreadsheets, so he learned SQL, then Python, then enough statistics to know when a chart is lying. At Datawheel he leads the team that builds forecasting and churn models for retail clients.
What learners say
4.9
1,090 ratings
Dina Mostafa
7 months ago
The data leakage lesson caught a mistake I'd made for a year: scaling before the train/test split. My accuracy dropped, and for the first time it meant something.
Rafael Costa
4 months ago
Building a churn model with a scikit-learn Pipeline, then explaining precision versus recall to an imaginary marketing team, is the most practical ML exercise I've done.
Mei Lin
2 months ago
Excellent intuition for decision trees, regularization and cross-validation. Neural networks only get a brief look, so don't come expecting deep learning.