Comparison of deepfake detection using CNN and hybrid models

In today’s digital era, manipulated videos known as deepfakes are becoming more common and harder to detect. These deepfake contents can be misused by people to spread false information, damage someone’s reputation, or even harm many people. To address this issue, we conducted a study to compare the performance of three deep learning models Convolutional Neural Network (CNN), Recurrent Neural Network (RNN), and a hybrid CNN-LSTM model to detect deepfake content. In this study, we used the DeepFake Detection Challenge (DFDC) dataset, which includes a variety of real and fake videos, then we will evaluate each model using accuracy, precision, recall, F1-score, AUC, and loss values. The results of our research show that the RNN model provides the best overall performance with an accuracy of 77.5%, F1-score of 77.5%, and AUC of 0.928, while CNN followed closely with 75.5% accuracy and AUC of 0.930. The CNN-LSTM model showed lower performance with 67.5% accuracy and AUC of 0.886. The findings demonstrate the strength of RNNs in modeling temporal patterns in video data, which is crucial for effective deepfake detection. Then CNN model also have good performance and shows strong results in detecting spatial features in individual frames. Meanwhile, CNN-LSTM model does not perform as well as the other models. This can be happen due to hybrid model having more complex structure, which makes it more difficult to train effectively. We hope this research helps in building better tools to detect deepfakes in real-world situations.