Data Balancing through Data Augmentation to Improve Transfer Learning Performance for Skin Disease Prediction