What is the acceptable minimum accuracy in image classification?

• A. 80%
• B. 75%
• C. 85%
• D. 90%

Answer: (C)

In image classification, the minimum accuracy you consider acceptable is a practical threshold that signals the model is reliable enough for use. There isn’t a universal rule, because what’s acceptable depends on how costly mistakes are in the real task. A common, sensible target many practitioners aim for is around 85%. This level shows strong, consistent labeling across most images and classes while still being attainable with real-world data and models. It provides confidence for deployment without demanding the very highest possible performance, which can be hard to achieve in noisy imagery.

If you dropped to 80% or 75%, misclassifications would be more frequent and could undermine decisions based on the results. Pushing toward 90% is great, but it’s often not practical as a default minimum for many applications. So 85% serves as a balanced, widely used benchmark for when an image classifier is ready to be trusted in practice, with the caveat that you should also check class-specific performance and the consequences of errors for your particular use case.