Leveraging AI for Sustainable Deepfake Human Face Detection Using Transfer Learning Technique

News and live streams on social media can have a significant impact on our daily lives. In a matter of seconds, news can reach audiences worldwide, influencing the economy, politics, or individuals either positively or negatively. Nowadays, deepfake has become a hot research topic due to significant advancements in artificial intelligence, particularly in artificial intelligence (AI). People are increasingly unable to recognize between genuine and fake images, voices, and videos. The material generated by generative AI is highly realistic and completely unique. Therefore, using AI technology to detect fake content is crucial. This study applied transfer learning to four models: VGG16, VGG19, MobileNetV2, and ResNet50V2. This article proposed a new model based on VGG16, where we concatenated the input from the pre-trained VGG16 model with the output from two additional convolutional and MaxPooling layers for further feature extraction, followed by three dense layers for deeper learning. The experimental values stated that the proposed model attains significant performance over other approaches with an accuracy rate of 98.4%. The findings also show that deepfake detection may be performed well in the future by leveraging cutting-edge machine learning techniques. This should lead to more advancements in the accuracy and robustness of deepfake technology detection and, therefore, a safer and more sustainable digital future.