I am training a deep learning model on imbalanced data for binary classification. I used binary_crossentropy for the loss function and Accuracy for the metric. When I plotted the loss, I got an underfitting. Is that a problem as my data is imbalanced and the accuracy does not reflect how well the model is?
Asked
Active
Viewed 72 times
0
-
If data is imbalanced, then accuracy is obviously not a correct metric to evaluate the model. You need to use Precision or Recall based on your task. If you bother about both precision and recall, you can consider F-score metric too, which is a harmonic means of Precision and Recall. – Ashish Jan 09 '22 at 07:19
-
Thank you for replying, I understand that accuracy is not used for evaluating the model in case of imbalanced data. Although my model performs well in terms of F1-score I am not able to interpret the cause of underfitting when I plotted the loss. – Siasma Jan 09 '22 at 20:32