COMPEL Glossary / accuracy
Accuracy
Accuracy is a model performance metric measuring the proportion of all predictions (both positive and negative) that are correct.
What this means in practice
While intuitive and commonly reported, accuracy can be severely misleading for imbalanced datasets. For example, if only 1% of transactions are fraudulent, a model that simply labels everything as 'not fraud' achieves 99% accuracy while catching zero fraud -- a useless result. For this reason, accuracy should never be the sole performance metric for AI systems. It should be reported alongside precision, recall, F1 score, and domain-specific metrics. In the COMPEL framework, model performance evaluation during the Evaluate stage requires multiple metrics appropriate to the use case, not just accuracy.
Why it matters
Relying on accuracy as the sole performance metric can create a false sense of confidence, particularly with imbalanced datasets where a model that predicts nothing achieves misleadingly high scores. Organizations that evaluate AI models using multiple metrics make better deployment decisions and avoid costly production failures. Regulators and auditors increasingly expect multi-metric evaluation as evidence of responsible AI governance.
How COMPEL uses it
During the Evaluate stage, COMPEL requires that model performance be assessed using multiple metrics appropriate to the use case — not just accuracy alone. This requirement is built into the gate criteria for production deployment. The Technology pillar assessment during Calibrate evaluates whether the organization has the monitoring infrastructure to track diverse performance metrics, and the Governance pillar ensures evaluation standards are codified in policy.
Related Terms
Other glossary terms mentioned in this entry's definition and context.