Quick answer

You need three areas of math for machine learning, and you need far less of each than most “prerequisite” lists suggest: basic linear algebra (vectors and matrices, since that’s literally how data is represented), enough statistics to understand distributions and what a model’s error actually measures, and just enough calculus to understand that training involves finding a minimum, without needing to compute derivatives by hand. Khan Academy covers all three for free from the ground up, and StatQuest on YouTube is widely recommended for explaining the statistics specifically in a genuinely intuitive way rather than pure formulas.

Math is the single biggest thing that scares people away from starting machine learning, and it’s also one of the most overstated prerequisites in the field. You do not need a math degree. You need a working intuition for a fairly small set of ideas, and you can build that intuition with free resources designed for exactly this purpose.

The three areas, and how much you actually need

  • Linear algebra: vectors, matrices, and basic operations on them. This matters because data in machine learning is represented as vectors and matrices, so understanding this is really understanding how your data is structured, not abstract math for its own sake.
  • Statistics and probability: distributions, mean and variance, and what it means for a model’s predictions to be “close” to the real answer. This is the math behind evaluating whether a model is actually good.
  • Calculus, in small doses: understanding that training a model means finding the settings that minimize error, which is fundamentally a calculus idea, without needing to compute every derivative yourself by hand. Libraries handle the actual computation.
These three areas connect directly to specific, concrete parts of how machine learning actually works, not abstract theory.

Where to actually learn each one for free

Khan Academy is the strongest single free resource here because it covers linear algebra, statistics and probability, and calculus from true beginner level, with short videos and practice problems, all completely free with no account paywall. It’s not built specifically for machine learning, so you won’t get ML examples directly inside it, but the fundamentals it teaches map directly onto what you’ll need.

  • Khan Academy: free, structured, ground-up coverage of all three math areas with practice problems, the most complete single free resource for the underlying math itself.
  • StatQuest (YouTube): widely recommended specifically for making statistics and machine learning math genuinely intuitive rather than just formula-heavy, a strong companion once you have Khan Academy’s basics.
  • 3Blue1Brown’s linear algebra series (YouTube): a well known, visually driven explanation of linear algebra that builds real geometric intuition rather than rote formula memorization.
Building basic fluency in each area before diving into machine learning theory makes the theory itself click faster, not slower.

A practical order to learn this in

You don’t need to master all three areas before touching machine learning at all. A workable approach is to start applied machine learning (Kaggle Learn, for example) immediately, and go back to Khan Academy or StatQuest specifically when a concept in your ML course doesn’t make sense yet. Learning the math attached to a real question you’re already trying to answer sticks better than learning it in the abstract, months before you’ll use it.

  • Start applied ML practice early, don’t wait for math mastery first.
  • When a concept doesn’t click (like why models overfit, or what a loss function is doing), go learn the specific math behind it.
  • Revisit linear algebra specifically once you start deep learning, since it becomes more central there than in basic ML.
Learning the math behind a specific concept when you actually need it tends to stick better than front-loading months of theory.
Khan Academy for structured fundamentals, StatQuest and 3Blue1Brown for intuition, used together rather than any single one alone.

Where this fits with everything else you’re learning

This math foundation supports everything else in machine learning rather than being a separate track. Once you’re comfortable with it, our guide to free supervised learning courses and free machine learning courses for beginners will make a lot more intuitive sense, and specific parts of deep learning that lean heavily on linear algebra will feel less mysterious.

Ready to put this math to use? Browse today’s free machine learning courses.

Browse free courses

Frequently asked questions

How much math do I really need for machine learning?

Less than most prerequisite lists suggest. You need basic linear algebra (vectors and matrices), enough statistics to understand distributions and model error, and just enough calculus to understand that training finds a minimum, without needing to compute every derivative by hand.

What’s the best free resource to learn the math behind machine learning?

Khan Academy covers linear algebra, statistics and probability, and calculus for free from true beginner level with practice problems. StatQuest on YouTube is widely recommended as a companion specifically for building intuition around the statistics and ML-specific math.

Should I learn all the math before starting machine learning?

No. A more effective approach is to start applied machine learning practice early and go back to learn specific math concepts when they come up and don’t make sense yet. Math tied to a real question you’re trying to answer tends to stick better than math learned in the abstract.

Do I need calculus for machine learning?

You need to understand the idea behind it, that training a model means finding settings that minimize error, which is fundamentally a calculus concept. You don’t need to compute derivatives by hand for real projects, since machine learning libraries handle that computation for you.

Why does linear algebra matter so much for machine learning?

Data in machine learning is represented as vectors and matrices, so linear algebra is really about understanding how your data is structured and manipulated, not abstract math disconnected from the actual work. It becomes even more central once you move into deep learning.