if i used ANN in matlab and during training the network the MSE for training was very low but for testing and validation still above 20, does that acceptance or not ?
what can i do to reduce MSE ?
your test dataset will be far different from the training dataset....
I suggest you to use 'cross entropy'.
Training, testing validation results must be closer. For example, if training accuracy is 98% and testing accuracy is 70% then it is not acceptable. Tune the parameters to obtain training and testing accuracy, sensitivity and specificity almost the same.
I have not seen your data, so I don't know if 20 is high or low for MSE. A metric like accuracy is easier to judge (since it is between 0 and 100 %).
If your MSE is very different for training and validation, one possible reason is that your training and validation samples are very different. See if there are any big differences between them.
- Badges
- Science topic