Comparative Analysis Of Deep Learning for Robust Deepfake Detection

With Deepfake technology rapidly growing and its use on the rise, there is an increased demand for effectiveness of its detection. This paper performs a comparative analysis of 4 deep learning algorithms – Meso-4, MesoInception-4, Resnet50, DenseNet-121. We will investigate each algorithm’s architecture, performance validation and versatility. The indepth overview will shed some light on accuracy, precision, recall, F1 score, etc. The dataset will encompass the real and fake community videos for a well-rounded assess on each algorithm’s effectiveness. The results of the research will help to with the potential practical applications of the reviewed algorithms, and mention possible changes and improvements. The results of the research have shown both strengths and weaknesses of each algorithm, giving clarity as to the potential use in practice, and changes that could be made in future. The overall aim of the dissolution is to guide the reader through the process of understanding the difference between the 4 algorithms – Meso-4, MesoInception-4, Resnet-50 and DenseNet-121.