What to know before you read.
- Begin with the complete data-to-evaluation workflow before collecting algorithms.
- Build baselines and error analysis into every project from the beginning.
- A finished, reproducible project is a stronger milestone than completing another model tutorial.
Stage 1: Python and data handling
Learn functions, collections, files and notebooks, then practise loading, filtering, grouping and visualising data. Your milestone is a reproducible exploration that explains column meaning, missing values, unusual records and a clear prediction question. Do not begin model training until the target and unit of observation are explicit.
Stage 2: Statistics and evaluation
Develop intuition for distributions, sampling, correlation, leakage, train-validation-test separation and task-specific metrics. Learn what false positives and false negatives cost in the chosen context. Evaluation is not a final step; it defines what the project is trying to improve.
- Define the target without future information
- Create a simple baseline
- Choose a metric from consequences
- Protect the test set
- Inspect results across meaningful groups
Stage 3: Core models
Start with linear and tree-based methods for regression and classification. Compare rather than collect. For each model, explain the important assumptions, how complexity changes overfitting and why performance differs from the baseline. Add clustering only when an unsupervised question genuinely exists.
Stage 4: One complete project
Choose a bounded public dataset and write a project brief before coding. Build a clean pipeline, record experiments, inspect error examples and explain limitations. Package the result with a README, reproducible environment and a short decision recommendation rather than presenting only a notebook and score.
Stage 5: Add depth deliberately
After the first project, choose one gap: feature engineering, model interpretation, deployment, monitoring or deep learning. The Machine Learning Certificate of Specialisation offers a structured 30-hour route across 2 Months, with labs and a capstone that connect these stages.
Questions readers often ask.
How long does it take to learn machine learning?+
A focused foundation and first project can begin within weeks, but dependable judgement grows through repeated projects and feedback.
How much mathematics is required?+
Begin with statistics and linear-algebra intuition. Deepen the mathematics when it helps explain model behaviour.
Should I learn every algorithm?+
No. Learn a small set well enough to compare baselines, diagnose errors and explain trade-offs.
Roadmap