AI & Machine LearningData Science

Overfitting

Perfect practice scores can hide failure in the real world

A model aces every practice question. Then it meets the real world and falls apart. How can perfect be a warning?

You'll find out

  • Why 8 out of 8 can be a bad score
  • What a model learns that won't come back
  • How data scientists catch it in time

Why it matters

Every app that predicts, from spam filters to maps, is judged on data it never saw. Here's why.

About 3 min6 interactive slides · 5 questionsLesson 271 · Zone 31

Play this lesson

Free to play in your browser or on iPhone

More in this zone

  1. 272Number TheoryCicadas Count in PrimesSome insects emerge every 13 or 17 years
  2. 273Media PsychologyNext Episode in 5…Removing one click keeps people watching for hours
  3. 274Account SecuritySIM SwappingSteal someone's phone number and you can reset their accounts
  4. 275History of ScienceMeasuring Light With Jupiter's MoonsIn 1676 eclipses of a moon revealed light's speed
  5. 276Angular MomentumThe Spinning SkaterPulling mass inward turns a slow spin into a fast one
  6. 277Orbital MechanicsWhy Satellites Don't FallThey are falling; they're just moving sideways fast enough to keep missing
  7. 278Software DesignHow Undo RemembersApps remember your actions, not copies of every version
  8. 279Media StreamingThe Buffering CircleA small reserve of video hides an uneven connection