The term top-5 error rate refers method of benchmarking machine learning models in the ImageNet Large Scale Visual Recognition Competition.

The model is considered to have classified a given image correctly if the target label is one of the model’s top 5 predictions.