Unmasking the Illusion: Deepfake Detection through MesoNet

In today’s era of vast digital manipulation, the rise of deepfake technology poses a significant challenge to genuineness of multimedia content and introduces a profound risk to privacy, cybersecurity, and information integrity. Our research contributes to the ongoing discussion on deepfake detection, with a particular focus on assessing effectiveness of MesoNet model. It focuses on analysis of face micro-expressions and makes use of the Face2Face deepfake dataset, known for its adeptness in facial reenactment. Objectives of the research include evaluating MesoNet's efficacy by scrutinizing its performance across various parameters, fine-tuning the model for improved results, and gaining nuanced insights into its capabilities. Results reveal a notable advancement, with MesoNet achieving an accuracy of 90.4%, surpassing the previous 89.1%. Improved results after careful adjustment of activation functions and regularization parameters underscores the significance of hyperparameter optimization in deep learning models.