Computer vision researchers continue to find automated face expression identification to be an intriguing and difficult topic. Because people's expressions may change so much, recognizing them using machine learning methods is a challenging task. The use of DNN for the purpose of emotion classification in human face images is an example of deep learning, a relatively new subfield of machine learning. The challenges of facial expression categorization have led to an extensive usage of CNN. In terms of methodology, the study employs 918 images from the CK+ dataset and performs data pre-processing such as normalization, converting images into gray-scale format, and reducing the dimensionality. Enhancement of data is made in order to make the model training easier, after that, division of data is done for the training and testing data set. After classifying the emotion, the ResNet50 model is used with a new fully connected layer. Result analysis reveals high accuracy (97.7%), precision (92%), recall (92%), and F1 score (91%), showcasing the superior performance of the ResNet50 model. Comparative analysis against other DL models like VGG-19 and CNN further confirms the superiority of the ResNet50 model. This paper offers insights into future research areas and possible applications for facial expression detection systems, which advances the field overall.