Validation accuracy stuck at 0.5

#1

I have downloaded the dataset from https://www.dropbox.com/s/7rjw6oet4za01op/CovidDataset-20200427T133042Z-001.zip?dl=0

Tried changing the learning rate, dropout value but still the validation accuracy remains 50%.
Following is the link to my code:
https://colab.research.google.com/drive/192qFHWTF8nKzc0HUxgRFShAJxKtuPa7N?usp=sharing

Any insights on the reasons? Is it overfitting? Is it underfitting? Why?

#2

hey @ambika11 ,
Its currently overfitting the data itself in the first epoch only.
If you plot the graph after training for a number of epochs. You will see ,the loss graph is going down and validation loss will keep on increasing . This is the sign of overfitting.
TO correct it

  1. If you have less data , you need to work a lot on the model architecture and also use a bit higher learning rate . I see that you were using a very low learning rate.
  2. Need to work upon augmentations to create more data.
  3. Use a bit lower dropout values.

Not a perfect try , but a little bit. Have a Look at this code file.

I hope this helps you.
Thank You :slightly_smiling_face:.