Abstract

Abstract: Deepfake technology's explosive growth has made it difficult to verify the legitimacy of digital content and created security threats across a range of industries, including politics, social media, and business communications. In order to improve video authenticity, this study suggests a hybrid deepfake detection system that combines Long Short-Term Memory (LSTM) networks with AlexNet, a potent convolutional neural network (CNN). While LSTM networks evaluate temporal sequences and capture the dependencies between frames in video streams, the suggested method makes use of AlexNet's powerful feature extraction capabilities to examine spatial information within video frames. Combining these models results in an efficient deepfake detection system that can accurately identify deepfakes and is resistant to subtle alterations. To assess the performance, experiments were performed using publically accessible datasets. The findings show notable gains in detection accuracy, with the hybrid system outperforming conventional CNN and RNN-based techniques in terms of recall and precision rate. Additionally, the suggested system showed resistance to a variety of deepfake methods, such as voice synchronization and face reenactment. This study opens the door for sophisticated deepfake mitigation techniques in the public and private sectors by highlighting the significance of integrating spatial and temporal analysis for trustworthy video authentication. A scalable solution for real-time video analysis is provided by the combination of AlexNet and LSTM networks, which is essential for halting the spread of false information and boosting confidence in digital media. Keywords: Deepfake detection, AlexNet, LSTM networks, video authentication, hybrid model.