Effect on performance from class imbalance in a multiclass-classification setting?

[shashankg7]

Hi,

I am trying out this model on my custom dataset with the following frequency distribution of class labels :

7: 23849, 0: 15159, 1: 6445, 4: 5759, 5: 3969, 3: 3659, 2: 2845, 6: 492

I am getting ~65% accuracy on ~16K testing samples after training on the above mentioned dataset. Can class imbalance be one of the reason for this low accuracy?

I am using the model in it's original setting (assuming the best settings as reported in the paper).