Deepfake Detection: Leveraging InceptionResNetV2 and LSTM for Enhanced Accuracy
Aradhna Saini, Attiuttama, Sheenam Naaz, Anushka Shivhare, Gaurav Dhuriya, Neha Yadav · 2025
The antique adage "seeing is believing" is simply no longer actual, which has profound implications for a ramification of sides of our lives. Deepfakes are becoming less complicated and simpler to create as technology advances. In fact, with the right software program, you can do a part of it right out of your palm. it is challenging to pick out deepfakes. The capability of the human eye to differentiate among deepfakes has reduced. however, a few academics have tried to identify deepfake. Deepfakes are media that can be generated by synthetic intelligence (AI) using algorithms. Artificial intelligence algorithms are designed to understand the traits of both the supply and the target photo. Subsequently, the target image is positioned on the pinnacle of the supply photograph. Our goal is to pick out video deepfakes by way of utilizing deep getting-to-know neural networks which include InceptionResNetV2 and LSTM. With the usage of transfer learning and the pre-trained InceptionResNetV2 CNN for characteristic extraction and vector production, we were able to efficaciously assemble a deepfake detection version. The capabilities had been used to train the LSTM layer, and the confusion matrix that was produced offers us the accuracy of testing and validation. The corresponding version's accuracy for 20 and 40 epochs was 84.75% and 91.48%, respectively.