The Observation Theory EncyclopediaFrom TSKAboutBy kindBy chapterBy Lean fileLedgerProvenance

balanced accuracy concept

DefinitionThe mean of the per-class recalls, so that a large class cannot hide a small one. Equation 0.34.
ExampleRecalls of 0.90 on class A and 0.30 on class B give balanced accuracy 0.60, whatever the class sizes.
BookData Mining as Observation, draft 0.2, commit f3914f0; entry id balanced-accuracy, kind concept.
Statusno ledger row names this entry. Corrections: none recorded.
Defining equation

Book equation 0.34.

Assumptions and scope
  • The mean of the per-class recalls, between the worst class and the best. Plain accuracy is the size-weighted mean of the same recalls, so a class of 990 rows recalled perfectly and a class of 10 never recalled give accuracy 0.99 and balanced accuracy 0.5.
  • The book prefers the min over classes to the mean where a verdict is at stake, since a mean can still hide one failing class among many.
Prior artnone recorded
Evidencelean/DataMiningAsObservation/BalancedAccuracy.lean
Reviewednot yet reviewed; generated 2026-09-10 from records at the commits on the provenance page.
00.51stratumscoreclass Aclass Bclass Cbarmean passes, the minimum fails
The mean of the per-class recalls, so a large class cannot hide a small one.

Equation

Book equation 0.34.

\[\mathrm{BA}=\frac1K\sum_{k=1}^{K}\frac{\text{correct in class }k}{\text{rows in class }k}.\]

Conditions

Conditions are curated in entries.toml rather than read from a record.

Ledger

none

First stated

Chapter 0 section 0.14 of Data Mining as Observation, with the program’s use in the anti-hub category table, openvector-bench/results/R13_STAGE1_RESULTS.md.

Measurements

Where the book states it Numbers, as the book’s sources table records them Source
chapter 10 section 10.2 five kinds, balanced accuracy 0.611, 0.718, 0.684, the category table, per-category 0.57 to 0.82, the sweep, pigeonhole floor openvector-bench/results/R13_STAGE1_RESULT.md:40-75

Failures and corrections

none

Invariance envelope

none declared

Machine checked

lean/DataMiningAsObservation/BalancedAccuracy.lean, theorems min_le_balanced, balanced_le_max, hidden_class, at observation-data-mining f3914f0; what the check covers is stated in the book’s appendix C.

Used in

Data Mining as Observation chapters 0, 10.

Related

min-over-strata; anti-hub; abstention; harness.

See also

Book equations stated beside the entry’s terms, not defining it: 10.7.

Ledger rows that cite the entry’s records without naming it: NEG-11.

Sources-table rows that share a record with the entry without naming it: chapter 3 section 3.5, chapter 10 section 10.2, chapter 11 section 11.6.

Status

Generated 2026-09-10 by encyclopedia/generate.py; book at observation-data-mining f3914f0; the commit of every record is listed in the encyclopedia’s provenance.

← bag of wordsbar →