DESIGN AND IMPLEMENTATION OF DEEPFAKE DETECTION SYSTEM

• Main Author: Calvin, Christopher
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/58857
• Institution: Institut Teknologi Bandung

Technology developments in the field of machine learning has been impacting many aspects of today’s life. Those impacts could be either good or bad. For example, with machine learning, it is easier to fool people by manipulating videos to make the subjects within those videos appear as other people. Such videos are commonly known as deepfake videos. Deepfake videos usually are used only for entertainment purposes. But there are some cases where deepfake videos are used for spreading false information. Because of that, the goal of this research is to design and develop a deepfake detecting system based on the technology of artificial neural networks. This system utilized Convolutional Neural Networks and Long Short-Term Memory to analyse an input video for temporal inconsistencies. The implemented model will be evaluated using accuracy and loss metrics gained from testing and training the model. Accuracy of the model is gained from comparing the number of right and wrong predictions of the predictions with the actual data while the loss of a model describes how far the predictions of the model are to the actual data. The trained model can distinguish between deepfake and real videos with 70% accuracy and loss of 0.66. Based on the experiments, it is found that by using transfer learning method, the separated model, which is trained separately from the CNN block, performs better than the combined CNN and LSTM model.