Deepfake Detection Using Convolutional Neural Networks and LSTM Modelling
Vishal Manishbhai Patel, Sheshang Degadwala · International Journal of Scientific Research in Science and Technology · 2025
This study proposes an advanced deepfake detection framework that combines Convolutional Neural Networks (CNN) with Long Short-Term Memory (LSTM) networks to effectively identify manipulated video content. The CNN component is designed to extract detailed spatial features from individual video frames, capturing subtle visual cues indicative of tampering. Meanwhile, the LSTM module models temporal dependencies across sequential frames, enabling the detection system to analyze frame-to-frame variations and inconsistencies characteristic of deepfake videos. This hybrid architecture leverages the complementary strengths of CNNs and LSTMs to enhance classification accuracy beyond conventional single-model approaches. The proposed Adaptive-GAN system, evaluated on benchmark datasets, demonstrates superior performance with a generator loss of 0.035 and discriminator loss of 0.020, reflecting stable and robust training dynamics. It achieves an impressive 97% accuracy, precision, recall, and F1-score, underscoring its effectiveness in distinguishing real from manipulated content. These results indicate that integrating spatial and temporal feature extraction substantially improves detection reliability, making the framework well-suited for real-time applications in digital media forensics. By addressing challenges in deepfake identification, this research contributes to the development of trustworthy AI-driven tools that can safeguard information integrity and combat misinformation in increasingly complex multimedia environments.