Hybrid Deepfake Detection System: Leveraging AlexNet and LSTM Networks for Enhanced Video Authentication
Abhijit Bapurao Ghodake, Durgesh Bajirao Yewale, Tejal Dilip Katariya, Bhakti Sahebrao Khobare, Ashvini S. Shidore · 2025
The rapid proliferation of deepfake technology has presented significant challenges in ensuring the authenticity of digital media, posing security risks in various domains, including social media, politics, and corporate communications. This paper proposes a hybrid deepfake detection system that integrates AlexNet, a powerful convolutional neural network (CNN), and Long Short-Term Memory (LSTM) networks to enhance video authentication. The proposed system lever-ages AlexNet's strong capabilities in feature extraction to analyze spatial features within video frames, while LSTM networks process temporal sequences, capturing the dependencies across frames in video streams. By combining these models, our approach delivers an effective deepfake detection system that is robust to subtle manipulations and capable of identifying deepfakes with high precision. Experiments were conducted on publicly available datasets to evaluate the performance of the hybrid model against standalone models and other baseline approaches. The results indicate significant improvements in detection accuracy, with the hybrid system achieving a higher recall and precision rate compared to traditional CNN and RNN-based methods. Furthermore, the proposed system demonstrated resilience against a wide range of deepfake techniques, including facial reenactment and voice synchronization. This work highlights the importance of combining spatial and temporal analysis for reliable video authentication, paving the way for advanced deepfake mitigation strategies in both public and private sectors. The integration of AlexNet and LSTM networks offers a scalable solution for real-time video analysis, providing critical support in com-bating the spread of disinformation and enhancing trust in digital media.