Facial paralysis is a state when a person is unable do the movement of the muscles of one or both the sides of his/her face which means that a person is not able to move the affected side of the face. Facial paralysis can be unilateral or bilateral. Paralysis often happens due to brain stroke. Facial paralysis detection is beneficial because it detects asymmetry in the face using the non-paralyzed part of face. This paper presents work on detection of unilateral facial paralysis. The performance of different deep learning algorithms such as CNN, InceptionV3, VGG16, and ResNet50 for detection of facial paralysis is analyzed. The results are showcased in terms of training and validation accuracy. The experimentation is carried out and results are compared at various epochs. We got the best training accuracy for Inception V3, which is 99.33%, achieved at the 39th epoch and the test accuracy of 89.87%, that is obtained at the 25th epoch. The best training accuracy obtained for CNN algorithm is 98.80% at the 29th epoch and the test accuracy for CNN is 88.92%, obtained at the 35th epoch. There were no promising results for other Deep learning models, VGG16 and ResNet50.