Machine Learning Fundamentals: Models, Training, Inference, and Evaluation

A model lands on your desk reporting 92% accuracy and asking for the green light. The room nods. You can't tell whether that's spectacular, ordinary, or — given the class imbalance no one mentioned — actually worse than predicting the majority class. By the time it ships and quietly fails on a subgroup six months later, you wish you'd known what to ask. This course gives you the conceptual fluency to know what to ask, without learning to train models in code and without a math-heavy detour. Across thirty-eight lessons it walks the durable layer of machine learning: the lifecycle from problem framing to monitoring; the data discipline that decides what a model can possibly learn; the major model families and what each one is and isn't good for; the training mechanics that explain why models overfit or fail to generalize; and the evaluation metrics that turn into honest claims — or, in the wrong hands, into theater. A full phase goes to the work most ML curricula shortchange: the metric that fits the cost of the error you'd actually pay, the calibration check that decides whether "70% probability" means anything, the fairness definitions you cannot all satisfy at once, the privacy properties that anonymization does not deliver, and the craft of saying plainly where a model is going to fail. Library names will rotate; benchmark leaders will turn over; this course is built on the layer that doesn't. You'll leave able to read an ML claim the way a senior data scientist does — and to ask the question that turns a vendor demo into a real decision.