Sure it can, however machine learning is more of a number-crunching, blocking task which requires certain calculation between nodes. It is a fact, that JS is inherently bad at that kind of work. So you will be able to do a school-project, however I cannot see any real-life application with JS/Node.JS only.

If your project is based on Node.JS and you need some kind of interface, I recommend using the Node.JS C++ API or FFI bindings with a system programming language, like Rust, in order to get maximum performance and still be able to kick off certain tasks or feed data from your Node.JS program.

every language that is able to use math can be used ;) .... the question should be along the lines "what are the pitfalls/problems of using node for ML"

-> missing libraries / community

medium.com/javascript-scene/how-to-build-a-neuron…

as an example starting point.

And than there is statistics and certain equations and that's all there is to ML we're just able to compute and aggregate enough data to have a better "statistical precision" that's why it's booming....