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?