Machine Learning Foundations
Understand how models learn from data, make predictions, fail, and improve. This course gives you clear mental models before you move into applied ML tools or heavier mathematics.
Start course
Prepare with Python
Read lessons without an account. Sign in only to save continuation, completion, and learner badges. No certificate or formal credential is issued.
Use Python knowledge to understand how models learn
Python Foundations gives you the coding base. Machine Learning Foundations uses that base to make datasets, features, training, evaluation, and prediction feel concrete instead of mysterious.
- Python Foundations or equivalent required
- No advanced mathematics required
- Built around clear mental models
Understand how ML differs from traditional programming and why examples matter.
Reason about data, features, labels, models, and the training process.
Distinguish regression, classification, and clustering, and explain which kind of question each can address.
Choose suitable metrics, inspect errors, and recognise overfitting, leakage, and limits of prediction.
See how cleaning, features, and messy inputs shape model behaviour.
Keep preprocessing and model evaluation together so experiments can be rerun and compared honestly.
Consider subgroup performance, uncertainty, human oversight, drift, and when a model should not be used.
Relate ML foundations to neural networks, deep learning, and modern generative AI.
Ten modules from problem framing to monitored predictions
Machine learning needs more than an algorithm overview. The production syllabus gives separate attention to framing, data, baselines, validation, metrics, leakage, error analysis, fairness, and behaviour after deployment, with one coherent project running through the course.
Frame the problem before choosing a model
- Prediction, description, automation, and decision support
- Defining the target, unit of analysis, users, and success criteria
- When rules, statistics, or no model are better choices
- Project step: write a problem and harm statement
Understand the data-generating process
- Examples, features, labels, and sampling
- Numerical, categorical, text, and time-based data
- Missingness, proxies, imbalance, and measurement error
- Project step: audit a dataset before modelling
Regression and classification foundations
- Continuous values, classes, scores, and probabilities
- Linear and logistic models as interpretable baselines
- Loss functions and the intuition behind learning
- Practice: train baseline regression and classification models
Unsupervised learning and representation
- Clustering, similarity, and dimensionality reduction
- What an unsupervised pattern does and does not prove
- Scaling and distance in feature space
- Practice: explore structure without inventing labels or causes
Generalisation, splits, and leakage
- Training, validation, and test sets
- Overfitting, underfitting, and model complexity
- Cross-validation and time-aware splitting
- Practice: find target leakage and repair an invalid experiment
Metrics, thresholds, and error analysis
- MAE and RMSE for regression
- Confusion matrix, precision, recall, F1, ROC-AUC, and PR-AUC
- Thresholds, class imbalance, and the cost of different errors
- Project step: choose metrics tied to the original decision
Preprocessing and reproducible pipelines
- Imputation, scaling, and categorical encoding
- Feature engineering without contaminating evaluation
- Pipelines and repeatable transformations
- Practice: compare a clean pipeline with a leaky workflow
Model families and honest comparison
- Bounded comparison of simple baselines, linear models, nearest neighbours, trees, and light ensembles
- Performance, complexity, interpretability, and maintenance trade-offs
- Hyperparameters, search, and the limits of leaderboard thinking
- Bridge: deeper algorithms belong in a follow-on course
Responsible ML and behaviour in use
- Subgroup performance, bias, fairness, and accessibility
- Interpretability, uncertainty, human oversight, and contestability
- Distribution shift, drift, feedback loops, and monitoring
- Practice: define deployment checks and a stop condition
Capstone and bridge to modern AI
- Complete an end-to-end supervised learning project
- Document data limits, baselines, metrics, errors, and risks
- Present conclusions without overstating causality or certainty
- Bridge: where neural networks, embeddings, and generative models fit next
Technical foundations without unnecessary intimidation
Learners who completed Python Foundations, developers and analysts adding ML, and anyone moving from AI use to AI understanding.
Basic Python is required; Python Foundations or equivalent experience is sufficient. No advanced mathematics is required for this foundation course.
Self-paced online lessons with hands-on model-building activities, clear conceptual explanations, browser notebooks, and an expected active workload of 36-44 hours.
Make sure the foundation is in place
If Python still feels unfamiliar, complete Python Foundations first. If you are entirely new to AI, start with AI Foundations before moving into code and models.
Course status: available now
The course is open as a free, self-paced learning path with text lessons, browser notebooks, downloadable files, account-saved progress, learner badges, and a completion checklist. It is educational material, not enrolment, assessment, certification, or professional advice.