A machine learning roadmap for beginners runs in six stages: (1) Python and data handling, (2) the essential maths (statistics, linear algebra, calculus intuition), (3) classical ML algorithms, (4) model evaluation, (5) deep learning and modern AI, (6) projects and deployment basics. At about 10 hours a week, most beginners reach an applied foundation in 9–15 months. Classical ML comes before deep learning, and evaluation skills matter as much as algorithms.

Key takeaways

• Python and data skills are the base layer. Don’t skip them.
• Learn maths alongside algorithms, not years before.
• Evaluation (train/test splits, metrics, overfitting) separates real ML from demos.
• Projects on real data prove skill better than course certificates.

The roadmap at a glance

Stage	Learn	Approx. time (10 hrs/week)
1	Python, pandas, NumPy, plotting, SQL basics	2–3 months
2	Statistics and probability; linear algebra and calculus intuition	2–3 months (overlaps)
3	Regression, classification, trees, ensembles, clustering	2–3 months
4	Validation, metrics, overfitting, feature engineering	1–2 months (overlaps)
5	Neural networks, transformers, using LLMs via APIs	2–3 months
6	Projects, version control, basic deployment	ongoing

Stage 1: Python and data

Get comfortable with Python syntax, pandas for tables, NumPy for arrays and matplotlib for charts. SQL is useful because real data lives in databases; see SQL for data analysis. If you’re coming from spreadsheets, read Python vs Excel for data analysis.

Stage 2: the maths you actually need

• Statistics and probability: distributions, mean/variance, conditional probability, hypothesis testing.
• Linear algebra: vectors, matrices, dot products, how data becomes numbers.
• Calculus intuition: what a gradient is, and how gradient descent minimises error.

Learn these in context: when you meet linear regression, learn the matrix form; when you meet training, learn gradients.

Stage 3: classical machine learning

Core algorithms, in a sensible order:

1. Linear and logistic regression
2. k-nearest neighbours
3. Decision trees, then random forests and gradient boosting
4. Clustering (k-means) and dimensionality reduction (PCA)

Use scikit-learn; its consistent interface lets you focus on concepts.

Stage 4: evaluation, the skill most beginners skip

• Train/validation/test splits and cross-validation
• Metrics: accuracy, precision, recall, F1, ROC-AUC, RMSE, chosen to fit the business problem
• Overfitting vs underfitting; regularisation
• Data leakage: the silent killer of impressive-looking results

Stage 5: deep learning and modern AI

Neural network basics, then PyTorch or TensorFlow, then transformers and how large language models work. Many applied roles now involve using foundation models via APIs, with retrieval, evaluation and safety, rather than training models from scratch. Practical prompting is covered in prompt engineering basics.

Stage 6: projects

Three projects that show range:

1. A tabular prediction problem (churn, prices) with honest evaluation
2. An NLP or computer vision project using a pre-trained model
3. An end-to-end app: a model behind a simple API or web interface

Is this the right path for you?

If you want to work with data but not necessarily build models, the analyst route is faster; see data analyst vs data scientist and how to become a data analyst. If you want AI work without coding, see AI jobs without coding.

Learn ML with a mentor

Tutopiya is planning a 12-week live AI & Machine Learning Essentials cohort (intermediate level, with some Python expected), with mentor-reviewed projects and a capstone. It isn’t running yet. Join the machine learning waitlist. Nothing is charged.

Sources

• scikit-learn user guide
• PyTorch tutorials
• Google: Machine Learning Crash Course

Frequently asked questions

How long does it take to learn machine learning? +

With consistent study of around 10 hours a week, many beginners reach a solid applied foundation (Python, core algorithms, evaluation and a few projects) in about 9 to 15 months. Depth for research or senior roles takes years.

How much maths do I need for machine learning? +

For applied work: linear algebra basics (vectors, matrices), calculus intuition (gradients), probability and statistics. You need to understand what algorithms are doing, not to prove theorems.

Should I learn machine learning or deep learning first? +

Classical machine learning first: regression, trees and evaluation. It teaches the fundamentals of overfitting, validation and feature engineering that deep learning relies on.

Do I need a degree to work in machine learning? +

Many ML roles prefer quantitative degrees. Applied and engineering-focused roles increasingly hire on demonstrated skill and projects, but competition is strong, so a robust portfolio matters.

Founding cohorts · nothing charged

Want to learn this live, with a mentor?

Tutopiya is planning small, mentor-led online cohorts in English test prep, data & AI, business & finance and digital careers. Tell us which one you want. The programmes with the most interest open first, and the people on the waitlist hear about them first.

See the planned programmes