It's such a broad cross-disciplinary area, it's tough to define a good entry point that is reasonable for everyone. But I'd say actually start with getting a so

It's such a broad cross-disciplinary area, it's tough to define a good entry point that is reasonable for everyone. But I'd say actually start with getting a solid foundation on statistics (point estimation, and hypothesis testing). If you're good with that, I'd learn a bit of DSP to get a feeling for how people in that world manipulate and clean digital signals. Some of that helps out later. FFT, DWT, PCA, ICA, Convolutional filtering, Kalman and PID filtering (if you're interested in online systems), things like that. They help especially in feature extraction for classification, but also for unsupervised methods. Past that, the hierarchy of methods to be learned and in what order is funny, I'm not sure there is a current consensus on that. The truth is that the best path kind of depends on your own goals. If you have a particular problem in your pocket that you want to solve, that actually helps a ton.