Applied Machine Learning Algorithms
Go beyond foundations and learn how to choose, compare, tune, inspect, reject, and explain practical machine-learning algorithm families with honest evidence.
Start course
Review prerequisite
Start immediately without an account. Progress is browser-local; notebook files are provided in the downloadable course pack.
Turn model-family names into better algorithm decisions
Machine Learning Foundations teaches the workflow: frame, audit data, split correctly, build baselines, evaluate, and state limits. This course uses that discipline to compare specific algorithm families without treating any method as universally best.
- Machine Learning Foundations or equivalent required
- Basic Python and notebooks required
- No advanced mathematics required
Choose model families based on data shape, sample size, feature type, decision cost, and explanation needs.
Use dummy, simple, interpretable, and complex candidates without turning the process into leaderboard chasing.
Record bounded tuning choices for regularised models, SVMs, trees, ensembles, and thresholds.
Use feature selection, permutation importance, partial dependence, and ICE-style thinking without claiming causality.
Recognise multiclass, multilabel, imbalance, anomaly, novelty, and semi-supervised boundaries.
Produce an algorithm portfolio and selection memo with clear keep, delay, reject, and limitation decisions.
Ten modules from candidate selection to final algorithm portfolio
The course is source-first and text-first. Each module includes lesson pages, an activity, an assessment, a notebook in the download pack, and a limitation-focused evidence task.
Algorithm Selection as a Design Choice
- Problem fit before algorithm names
- Data size, feature type, noise, missingness, and dimensionality
- Baseline ladder and comparison protocol
- Project step: algorithm candidate plan
Regularised Linear and Logistic Models
- Linear models as strong baselines
- Scaling, coefficients, ridge, lasso, and Elastic Net
- Logistic regression, class weights, and threshold caution
- Project step: regularised model comparison
Distance-Based Methods and Kernel Intuition
- What distance means in feature space
- k-nearest neighbours and scaling
- Dimensionality, irrelevant features, and kernel intuition
- Project step: distance-sensitive model report
Probabilistic Classifiers and Calibration
- Scores versus reliable probabilities
- Naive Bayes variants and assumptions
- Calibration curves, calibrated classifiers, and thresholds
- Project step: probability-quality report
Support Vector Machines
- Margins and support vectors in plain language
- Linear SVMs, SVR, kernels, C, and gamma
- Runtime and explanation trade-offs
- Project step: SVM candidate review
Trees, Pruning, and Rule-Like Models
- Splits, leaves, impurity, and depth
- Classification and regression trees
- Overfitting, pruning, missing values, and categorical caution
- Project step: pruned tree report
Ensembles: Bagging, Forests, Boosting, Voting, and Stacking
- Bias, variance, diversity, and bagging
- Random forests and gradient boosting
- Voting, stacking, early stopping, and leakage-safe tuning
- Project step: ensemble comparison
Feature Selection, Dimensionality, and Inspection
- Low-variance, univariate, recursive, and model-based selection
- Pipeline-safe feature selection
- PCA, permutation importance, partial dependence, and ICE limits
- Project step: feature and inspection memo
Special Problem Settings
- Multiclass, multilabel, and multioutput problems
- Imbalanced data beyond simple class weights
- Anomaly, novelty, and semi-supervised boundaries
- Project step: choose or reject a special-setting method
Capstone: Algorithm Portfolio and Selection Memo
- Confirm the problem, data, split, and baseline
- Build and tune a bounded candidate portfolio
- Compare performance, complexity, calibration, stability, and inspection evidence
- Write the final algorithm choice memo
Applied algorithm depth with clear limits
Learners who can already complete a basic machine-learning workflow and now want better judgement about model families.
Machine Learning Foundations or equivalent skill: train/test discipline, baselines, metrics, pipelines, and responsible-use limits.
Self-paced online lessons with a downloadable pack of notebooks, datasets, helper code, and templates. Expected active workload is 40-55 hours.
Make sure the workflow habits are already in place
This course assumes you already know how to frame a problem, inspect data, protect evaluation, build baselines, and explain model limits. If those ideas feel shaky, complete Machine Learning Foundations first.
Start Applied ML Algorithms
Review Machine Learning Foundations
Course status: available now
The course is open as a free, account-free, source-first learning path with text lessons, downloadable notebooks, synthetic datasets, helper code, local progress, and a completion checklist. It is educational material, not enrolment, assessment, certification, production approval, or professional advice.