How can i solve overfitting?

TensorFlow General Discussion
1

i got this results during training and i set epoch 100 … will the results can be change to be better or this indicates to be over fitting ?

1 loss: 4.5090 - accuracy: 0.2570 - val_loss: 3.8487 - val_accuracy: 0.3179
2 loss: 3.7380 - accuracy: 0.3306 - val_loss: 3.6578 - val_accuracy: 0.3398
3 loss: 3.5954 - accuracy: 0.3449 - val_loss: 3.6016 - val_accuracy: 0.3481
4 loss: 3.5336 - accuracy: 0.3519 - val_loss: 3.5808 - val_accuracy: 0.3527
5 loss: 3.4969 - accuracy: 0.3563 - val_loss: 3.5751 - val_accuracy: 0.3555
6 loss: 3.4720 - accuracy: 0.3595 - val_loss: 3.5733 - val_accuracy: 0.3574
7 loss: 3.4528 - accuracy: 0.3619 - val_loss: 3.5809 - val_accuracy: 0.3585
8 loss: 3.4388 - accuracy: 0.3638 - val_loss: 3.5888 - val_accuracy: 0.3595
9 loss: 3.4283 - accuracy: 0.3655 - val_loss: 3.5965 - val_accuracy: 0.3606
10 loss: 3.4207 - accuracy: 0.3666 - val_loss: 3.6070 - val_accuracy: 0.3612
11 loss: 3.4155 - accuracy: 0.3675 - val_loss: 3.6225 - val_accuracy: 0.3615
12 loss: 3.4112 - accuracy: 0.3688 - val_loss: 3.6362 - val_accuracy: 0.3622
13 loss: 3.4085 - accuracy: 0.3696 - val_loss: 3.6467 - val_accuracy: 0.3623
14 loss: 3.4024 - accuracy: 0.3697 - val_loss: 3.6510 - val_accuracy: 0.3617
15 loss: 3.3850 - accuracy: 0.3702 - val_loss: 3.6485 - val_accuracy: 0.3622
16 loss: 3.3729 - accuracy: 0.3709 - val_loss: 3.6563 - val_accuracy: 0.3622

can i share the model here ?

3

what is your model and input shape

5

This is a little bit hard to say without more information.
This could show that your model is not converging to a solution as the accuracy are not moving much, but that could also be specific to your use case. Maybe your model is too simple and cannot extract the complexity of your data.

To know if your model is overfitting, the easiest way is to finish your training and then evaluate your model on your dataset’s test split. If the results are much worse than the training accuracy, that’s a good indicator of overfitting.

some more resources here: Overfit and underfit  |  TensorFlow Core

2 Likes