Same training data in different projects give different confidence scores

I trained the same dataset in two projects and when I do the query search I get different confidence levels from intents. Can anyone explain why this happens? And in the same project already with a model, if I train the same model without changes in the dataset I get different scores of confidence for the same intents. Im using “intent_classifier_tensorflow_embedding”

How different are the confidence?

Like 0.85 and 0.75 Or 0.85 and 0.50

Like 0.85 and 0.75. I already discovered that it was of the random_seed parameter of the intent_classifier_tensorflow_embedding

Better way to make sure is to evaluate and check the f1 score to ensure the integrity of the model. Accuracy is not a great metric