There is a particular kind of machine learning course that respects your intelligence enough to make you work, and this is one of the clearest examples of it. It sits inside MIT's MicroMasters in Statistics and Data Science, and it carries the institution's house style, which is to refuse to let you skate over the mathematics. Where a typical applied course hands you a fitted model and a satisfied feeling, this one walks you back to the linear classifier, makes you understand the loss and the gradient, and then has you implement the thing in Python so that the understanding is not theoretical but lived. The arc of the syllabus is genuinely broad, starting with linear models and support vector ideas, moving through neural networks and deep learning, touching kernels, unsupervised methods and clustering, and finishing with a stretch on reinforcement learning, and the projects are where the real teaching happens, because they are large enough to force you to confront the gaps in your understanding rather than paper over them.
That is the case for taking it, and it is a strong one. The honest warnings are about fit and effort rather than quality. The mathematical prerequisites are not a polite suggestion, you need comfort with linear algebra, calculus and probability, and learners who arrive hoping the course will gently supply that background tend to stall in the first few weeks feeling that everyone else got a manual they did not. The pace is firm and the difficulty is real, which is exactly what makes the course worthwhile but also what makes it the wrong first stop for someone who only wants an applied overview.
Support is lighter than you would get inside a paid live cohort, so you are relying on the discussion forums, your own stubbornness, and a willingness to sit with a hard idea until it yields. And the certificate, while it exists and costs money on the verified track, is close to irrelevant next to the actual prize, which is the understanding itself. My view is that this belongs near the top of any list of free machine learning courses for the specific person it was built for, someone with the mathematical footing and the patience who wants to come out the other side able to reason about these methods from first principles rather than recite which function to call. If that is not you yet, there is no shame in building the prerequisites first or starting somewhere gentler, because taken at the wrong moment even an excellent course can simply convince a capable person that they are not cut out for the field, which would be exactly the wrong lesson to draw from it.