Metric & Benchmark
Accuracy
Fraction of predictions that match the ground truth label, the simplest classification metric.
Definition
Accuracy is the share of correct predictions over total predictions. It is intuitive and widely reported but misleading on imbalanced datasets â a 99%-negative dataset trivially scores 99% by always predicting negative. Use precision/recall/F1 alongside it.
Common use cases
- Classification benchmarks
- Multiple-choice eval
- Sanity checks