DeepFake Video Detection Using Hybrid ResNet18-BiLSTM Architecture with Spatiotemporal Learning and Real-Time Forensic Analysis

This research presents a hybrid deepfake video detection framework integrating ResNet18 and BiLSTM architectures for robust spatiotemporal forensic analysis. The proposed system combines spatial feature extraction and temporal sequence modeling to detect manipulated videos with high accuracy. Experimental evaluation was conducted using FaceForensics++ and CelebDF datasets. The proposed model achieved 96.30% accuracy and 0.9922 AUC score, demonstrating strong robustness and generalization capability for real-time deepfake detection applications in cybersecurity and multimedia forensics.