Deep learning, a neural network model built with keras framework, training the given accuracy of 0.82, but in the 1500 test sample testing accuracy is only 0.33, don't know is why?
Think about a possibility, training set of sample size is not uniform, some training sample more than 1, 500, have only 5 pieces,
If the training sample uneven will cause this kind of situation?

CodePudding user response:

Training should see the accuracy of test set at the same time, I feel you say that big probability is overfitting

CodePudding user response:

Class imbalance may be the cause of this problem, but most of the unbalanced data rather than accuracy, at this moment need other comprehensive index value, such as F1 G - means value to measure the performance of the model, so, suggest you consider whether this phenomena caused by fitting,