MENTAL MODEL · M0769

Accuracy Paradox

TechnicalHigh supportStatistics

Included

Version 1.0.0 · Updated 2026-07-30

CORE DEFINITION

In cases of extreme class imbalance (e.g., 99% good people, 1% bad people), a model that predicts 'good' for all instances achieves 99% accuracy, but this model is completely useless (because it fails to catch the bad ones).

SCAFFOLDING EFFECT

psychology

Reduce cognitive load

Beware of vanity metrics. When you hear '99% accuracy', don't rush to applaud; check the baseline probability. For rare event prediction (e.g., cancer screening, fraud detection), recall is often more important than accuracy.

anchor

Anchor fast decisions

Under imbalanced data, accuracy = correct predictions / total, dominated by the majority class. A 'dumb model' that predicts the majority class for all instances has inflated accuracy but no discriminative power. What really matters is the identification of the minority class (recall, precision, AUC).

MINIMUM ACTION

In progress 0/5

Practice this model in one real situation:

First check the baseline class distribution (majority class proportion).
Don't rely solely on accuracy
use a confusion matrix.
For imbalanced problems, focus on Recall / F1 / AUC.
If necessary, resample or change evaluation metrics.

Check to track your progress (stored locally)

Learning progress0%

account_treeGenealogyexpand_more

menu_bookReferencesexpand_more

Source support: Explicit

link
en.wikipedia.orghttps://en.wikipedia.org/wiki/Accuracy_paradoxZH · Explicit
verified

RELATED MODELS