Title: Unlocking the potential of deepfake generation and detection with a hybrid approach

Authors: Shourya Chambial; Tanisha Pandey; Rishabh Budhia; Balakrushna Tripathy; Anurag Tripathy

Addresses: Vellore Institute of Technology Vellore, VIT, Vellore Campus, Tiruvalam Rd, Katpadi, Vellore, Tamil Nadu 632014, India ' Vellore Institute of Technology Vellore, VIT, Vellore Campus, Tiruvalam Rd, Katpadi, Vellore, Tamil Nadu 632014, India ' Vellore Institute of Technology Vellore, VIT, Vellore Campus, Tiruvalam Rd, Katpadi, Vellore, Tamil Nadu 632014, India ' Vellore Institute of Technology Vellore, VIT, Vellore Campus, Tiruvalam Rd, Katpadi, Vellore, Tamil Nadu 632014, India ' Carnegie Mellon University, 5000 Forbes Ave, Pittsburgh, PA 15213, USA

Abstract: The rapid advancement of software and AI has led to the proliferation of deepfake media, raising ethical concerns due to its potential for manipulation and deception. We propose a hybrid model combining Inception ResNetV2 and Xception architectures, leveraging LSTM and CNN for precise deepfake detection. With a diverse dataset, our study aims to identify and mitigate the harmful effects of deepfake technology, including financial fraud and misinformation dissemination. Through rigorous training, our model achieves an impressive accuracy of over 96.75% in detecting deepfake videos. This represents a significant step forward in combating deceptive content and upholding authenticity in the digital sphere. Our research emphasises the societal importance of deepfake detection methods, contributing to the broader mission of fostering trust and accountability online.

Keywords: fake video; convolutional neural network; CNN; recurrent neural network; RNN.

DOI: 10.1504/IJCSE.2025.144802

International Journal of Computational Science and Engineering, 2025 Vol.28 No.2, pp.151 - 165

Received: 29 Apr 2023
Accepted: 27 Dec 2023
Published online: 03 Mar 2025 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article