I am using a classification model that classifies normal and abnormal in binary through imagedatagenerator. Training and validation generators have been performed and model training has performed well. However, a question arose in the test generator.
The difference in classification performance indices between shuffle=True and False in the test generator is too large.
When shuffle=False, it results in a very good classification model. shuffle=True gives 50% accuracy.
I've seen other users use the test generator, but I don't understand why shuffle is set to False. help?
I think the test accuracy is reliable because the data built in the test folder is also randomly shuffled when shuffling is performed.
