I'm training a CNN-LSTM concat model and after 20 epochs I got an accuracy of 69% and a loss of 0.04?
I know that the combination of very high training accuracy and relatively low validation accuracy indicates overfitting but I was wondering if a low accuracy and a very low loss would also indicate overfitting.
Overall, the accuracy increased linearly and the loss decreased exponentially.
Some more details regarding task, loss and so on would be appreciated.
For the training set the loss should have a inverse relationship with the metric e.g. accuracy. If the loss goes down the accuracy should go up. If this is not the case you might want to have a look at the loss function and metric.
So a very low loss and very low accuracy on the same dataset would indicate for me, that something is wrong with the model I want to train and that the model is not learning to solve the task I want it to solve.
