Loss vs Accuracy: Binary Classification

[Fatema080136]

Thanks, Daniel Bourke, for this great tutorial.

You challenged us to improve the accuracy of the binary classification model to 80% (12:33:03). I was able to achieve 100% accuracy (Epoch: 3900 | Loss: 0.10130, Accuracy: 100.00% | Test Loss: 0.13714, Test Accuracy: 100.00%).

During my experiments, I noticed that while accuracy was improving, the loss was not decreasing (e.g., Loss: 0.49138, Accuracy: 93.50% | Test Loss: 0.50832, Test Accuracy: 89.50%). I am wondering why this is happening. Could you please help me understand this?

I am also quite frustrated that the model gives different accuracy results on different runs.

[VishmaPratimDasHere]

Just some primary observation, maybe it'll help you find a solution in the future: Since your accuracy is a 100% on the train dataset, it is quite possible that your model is using unrelated parameters (or noise) to find the result. (To know more, google "Overfitting and Underfitting")
In your case it is overfitting.
If there is some way you can decrease this, maybe your model will give better performance on the testing dataset.

[Harshith-reddy-td]

It is typical overfitting issue. Answer to the different accuracy on different runs is because you didn't use randomseed( ) which splits test train same everytime.