Skip to main content
PProfileMatch
Read Play
← Path

Supervised Learning · Class Imbalance, Metric Choice and Data Leakage: Why Results Look Too Good

0 XP
  1. 1 current stop: learning
  2. 2 upcoming
  3. 3 upcoming
  4. 4 upcoming
  5. 5 upcoming

Finish after stop 5

Stop 1 of 5 · Parcel depot, sorting line

Trick to keep in mind+3 XP

Few flags, all hits: great precision, tiny recall. Count what slipped through.

Compare accuracy with the do-nothing score first: when 4% of parcels are damaged, saying 'fine' for every parcel is right about 96% of the time. A model that flags only a handful can have excellent precision and still miss nearly all the damage. Work out how many damaged parcels were found among all of them.

Read it once, then a question tests whether the trick stuck.