As digital forgery blurs the line between truth and illusion, deep-fake technology presents unique problems for the integrity and reliability of online media. The use of fakes that can mimic real videos has sparked concerns about how they can be abused to spread false information and erode public confidence in real ones. As a result, it is important to ensure the authenticity of digital content by effectively detecting manipulated videos. In this study, we performed several experiments on ResNet50 trained using the Celeb-DF dataset using different hyperparameters and fine tuning methods to analyze them on the basis of metrics and evaluate its performance. Of the 12 various experiments performed, the model performed exceptionally well in some hyperparameters, achieving an AUC score of 98.11%. The findings of this article attempt to help build confidence in society in light of the growing threat posed by deepfakes.