I am trying to determine the number of neurons in the hidden layer of the network, when the MSE value is low, the Regression is also low. Why is this happening? Shouldn't All regression value be low when MSE is low??
I suggest you follow :https://www.researchgate.net/post/What_is_the_Acceptable_MSE_value_and_Coefficient_of_determinationR2Article Comparison of Neural Network Error Measures for Simulation o...
https://arxiv.org/ftp/arxiv/papers/1312/1312.2853.pdf
Any loss function depicts the differences between your actual values and predicted ones. i.e the MSE value reflects how your model can find a line to distinguish between your input data.
Concerning this, the smaller MSE value shows more precise predictions. However, it does not necessarily lead to smaller regression parameters. In fact, models will find the appropriate hypothesis with any appropriate values!
To sum up, the answer to your question: NO, it should not be.
In general, No. However, when model is sensitive or over fit, this may happen. I prefer MSE, which is better than regression in many aspect. Good luck.
What do you mean with 'The regression is low'? Do you mean that your predicted values are low?
If so, you might want to do a few thing:
-Check how your training and validation loss behave throughout the training process. Make sure you don't have overfitting and that both curves are mostly decreasing.
-If your output values or input values are too large (i.e., above the thousands ), then you may want to consider normalizing either the input values, the output values, or both.
-Make sure you don't have an activation function in your last layer.
-As other people said before, you should also consider the RMSE value, but this is not as important as the other points I mentioned.
Hope this helps.
Cheers
- Badges
- Science topic