Accuracy
The degree to which an AI system's outputs correctly reflect real-world data or intended outcomes.
Definition
More than a single percentage score, accuracy must be measured across multiple dimensions: overall correctness (true positives + true negatives), subgroup performance (e.g., by region, demographic), and edge-case robustness (rare conditions). Only by analyzing these facets can organizations ensure the system behaves reliably in production and identify scenarios where additional training or model adjustments are required.
Real-World Example
An autonomous vehicle company tests its pedestrian-detection AI in sunny, rainy, and nighttime conditions. While the model is 98% accurate overall, it drops to 85% under heavy rain. Engineers then augment training data with rain-specific footage and install additional infrared sensors, boosting rainy-condition accuracy back above 95% before the next public rollout.
.svg)
