In this research, we focus on investigating and evaluating the effectiveness of different adversarial attacks and see how resilient ML and DL algorithms are in classifying encrypted traffic applications. We train our models in the adversarial-free environment on encrypted traffic datasets using the top five features. Then, we use adversarial samples to attack the models and check their resilience in classifying the encrypted applications. Finally, we retrain the models using the top nine features in the adversarial-free and adversarial environment to compare the effect of the adversarial samples in both scenarios. We demonstrate that DL shows better resilience against the adversarial samples in encrypted traffic classifications in comparison to ML. We also indicate that ML and DL model's performance in an adversarial-free environment and their resilience against adversarial samples was improved when using more top selected features. However, the impact of the attack varies depending on the type of attack.