20 QUESTIONS · 5 TOPICS · THRESHOLD SLIDER · ABOUT 12 MINUTES

A model that's 99% accurate can still be useless.

If 1 in 100 transactions is fraud, a model that always says "not fraud" scores 99% and catches nothing. Twenty questions on how machines learn and how to tell if they really have, then tune a spam filter yourself. No sign-up, nothing saved.

Start the test
THE "ALWAYS SAY NO" FRAUD MODEL
99%accuracy
0%of fraud caught (recall)
QUESTION 1 / 20 · CORE CONCEPTSEasy

Predicting a house price from its size and location is a…

PRECISION VS RECALL

Where should the spam filter draw the line?

A model gave 24 emails a spam score. Everything right of the line goes to the spam folder. Move it and watch what you catch, and what you wrongly bin.

threshold 50
score 0 · inboxspam folder · score 100
actually spamactually a real email
flagged spamleft in inboxspam9
caught
2
missed
real4
wrongly binned
9
correctly kept
69%Precisionof binned emails that really were spam
82%Recallof all spam that got caught
0.75F1 scorebalance of the two, 0 to 1
A balance. For spam, most people prefer fewer wrongly binned emails (high precision); for disease screening, catching every case (high recall) matters more.

What this machine learning test covers, in short

Twenty multiple-choice questions at introductory undergraduate level, four on each of five topics: core concepts, training and optimisation, evaluation, classic algorithms, and deep learning. They test how models learn from data and how to judge whether they work.

No coding required. The test is our own and not normed. For background, read the machine learning study guide.

By What's Your IQ · Reviewed 5 Oct 2026
Format
20 multiple-choice questions
Level
Introductory undergraduate
Bonus
Spam-filter threshold and confusion matrix
Term coined
Arthur Samuel, 1959

Keep going

Machines find patterns in data. How fast do you?

Our full IQ test covers pattern, logical, numerical and spatial reasoning: 60 questions, about 30 minutes.

Take the full IQ test