Accuracy
The degree to which an AI system's outputs correctly reflect real-world data or intended outcomes.
Définition
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.
Exemple concret
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)
