New: Roadmaps ordered paths through our cheat sheets and flashcards, so you always know what to study next.
Explore themSee what's new on GitHubFrom telling supervised and unsupervised apart to training, evaluating, and automating a model you can actually defend.
A 12-step learning path. Follow it in order, or jump to what you need.
For analysts, product people, and students who want to see how machine learning models actually get built and judged, not just use them. Plan on about 6 to 9 weeks at a few hours a week, moving from the four learning paradigms through feature engineering, ensembles, model evaluation, and a first real look at deep learning. This path stops at that theory and evaluation layer: it does not cover shipping, serving, or monitoring a model in production, the ground Machine Learning Engineer picks up from here. By the end you can pick the right learning paradigm for a dataset, train and validate an ensemble model you can trust, and automate a full pipeline from raw features to a judged result.
Expected: comfort reading data in a spreadsheet or table. Helpful but not required: basic Python and an intro statistics course.
Read this once and you can already tell whether a new dataset calls for supervised, unsupervised, or reinforcement learning, the split every later step in this path assumes you can make.
Takes the supervised half of step 1's split and shows you real algorithms, so you can match a labeled dataset to logistic regression, a tree, or an SVM instead of guessing.
Covers the half of step 1's split that has no labels to lean on: clustering and dimensionality reduction, named a fast-growing skill on Coursera's own 2026 enrollment numbers.
You can look at a new dataset and know whether you're clustering it, classifying it, or predicting a number, with working algorithms in hand for each case. Next up: turning raw columns into something a model can actually learn from.
Finish this section to unlock.
+100 XP
Turns the raw columns steps 2 and 3's algorithms need into numbers they can actually read: encoding, scaling, and the leakage traps that quietly wreck a model's test score.
Puts the algorithms from steps 2 and 3 and the features step 4 built into one consistent Python API, so you can train and compare several models in an afternoon.
Shows why Random Forest and gradient boosting, the algorithms step 5's API keeps defaulting you toward, usually beat any single model: combining diverse mistakes instead of trusting one guess.
This is where the bias-variance tension from step 2 stops being a phrase and starts deciding whether you trust a score; expect to keep circling back to cross-validation once real modeling gets messy.
You can engineer real features, train and combine several algorithm families into an ensemble, and back the resulting score with a validation strategy that isn't flattering you. A few minutes of due flashcards on steps 1 to 3 keeps that paradigm vocabulary sharp while you push further here. Next up: the rare cases your metrics still miss, and the models that think in layers.
Finish this section to unlock.
+100 XP
Neural networks are where the algorithms from step 5 stop being the default answer; give this one a second pass before the layers-and-weights picture clicks, it usually takes one.
Points steps 3 and 7's tools at a problem most business data actually has: the rare fraud case or defect that standard evaluation metrics quietly average away.
Take this if you're chasing the extra accuracy that wins Kaggle-style tabular competitions: the exact tuning levers behind the gradient boosting step 6 introduced only in outline.
Aim here if your job means defending a model's call to a boss or a regulator: the SHAP and LIME tools that turn step 8's black box back into a reason.
Pulls every earlier step into one pipeline: the features from step 4, the model families from steps 2, 3, 6, and 8, and the evaluation discipline from step 7, automated end to end so you can judge a system instead of hand-tuning one piece at a time.
You can frame a problem by paradigm, engineer honest features, train and validate real models including ensembles and a first neural network, and automate a full pipeline end to end. That's the Machine Learning Foundations badge, earned.
Finish this section to unlock.
+100 XP
Finish every required step, at least 70% of them genuinely done (not skipped), to earn this badge and 500 XP.