Model performance

[nandu-ncs]

With 1.6438261270523071 loss and 0.5646336674690247 accuracy, how could you get that predictions and confusion matrix so accurate.

[dinachoir]

The model performance is not good, because the training is only applied to 10% of the total training dataset. This section is only an experiment to show that with only 10% of the training dataset and the use of transfer learning, we can achieve 'not so bad' performance. In real life case, you will do many experiments to find the most effective model for your dataset, that's why you will train the model with 10% of total training dataset first. Training with 100% of training dataset is here:
https://github.com/dinachoir/Convolutional-Neural-Network/blob/main/Food%20Image%20Recognition.ipynb

[nandu-ncs]

So, may I know the model and parameters under which you got that confusion matrix.