Track 3 · Machine Learning Fundamentals

From analysing the past to predicting the future. This track demystifies machine learning by having you do it: train a real model, evaluate it honestly, and confront the responsibility that comes with it.

What you’ll be able to do

• Frame a prediction problem: features vs target
• Split data into training and test sets — and explain why
• Train a decision tree classifier with scikit-learn
• Read accuracy and a confusion matrix without fooling yourself
• Measure your model’s fairness with code — per-group error rates and the four-fifths rule — and place the findings correctly in EU law, which imposes duties but prescribes no fairness metric
• Spot overfitting, choose a model with cross-validation, and compare models on fairness as well as accuracy
• Write a model card and export a portfolio of your work

The dataset

Student Digital Skills — 900 learners with their study field, weekly online-learning hours, digital-skill score and device access; the target is whether they passed a digital-skills certification. (Synthetic, education-sector themed.)

Lessons

#	Lesson	Time
1	Train your first classifier	~90 min
2	Overfitting, and choosing a model	~90 min
3	Capstone: audit and document a model	~2 h

Fair warning

Every lesson ends with an ethics lab. You will measure your own model failing learners who can’t afford a laptop, discover that the standard fairness screen misses it, find that a more accurate model makes the gap wider, and put your deployment decisions in writing. Those are the most important cells in the track — and the part no other platform teaches.

The capstone collects those written decisions, plus your own model card, into a portfolio you can download and submit for formal assessment — see Earning a microcredential.