Clear explanations of key ideas in machine learning, one concept at a time.
Topics: All articles | Machine Learning Explained | Exam Preparation | Problems and Solutions | Studies and Careers | Competition News

What Is Data Leakage in Machine Learning? A Package Test
A package model can score perfectly by using the time the package actually arrived. At dispatch, that timestamp does not exist yet. Read more

Why Is Feature Scaling Important in KNN?
A bike ride can have a different nearest neighbor after scaling, even though nothing about either ride changes. Follow the distances by hand to see why. Read more

How Does a Decision Tree Work? Follow One Prediction
Follow one library return through a small decision tree. At each split, see which training examples stay in play and which ones drop out. Read more

How to Read a Confusion Matrix for a Spam Filter
A spam filter can block a real email or let junk through. Build a two by two table from eight predictions and see where each mistake belongs. Read more

Gradient Descent Explained Simply With Two Updates
Work through two gradient descent updates by hand. Then see how changing the learning rate can shorten a step, cross the minimum, or make the loss rise. Read more

What Is Overfitting in Machine Learning? A Tale of Two Curves
One curve hits every training point. The other misses all three but predicts two new points more accurately. Plot them both to see why. Read more