Deepfake Detection using AI – Manipulated Videos and Images

Artificial Intelligence (AI) has made it possible to develop deepfakes realistically manipulated videos and images, primarily created with Generative Adversarial Networks (GANs). Innovative though they are, their malicious use can lead to dangers in misinformation, political manipulation,fraud, and identity theft.This project introduces an AI-based deepfake detection system with Convolutional Neural Networks (CNNs),with XceptionNet being the baseline model, aided by preprocessing operations like face detection, frame extraction, and normalization. Trained on benchmark dataset including FaceForensics++ and DFDC, the system obtained 92.3% accuracy with robust precision, recall, and AUC-ROC values. Grad-CAM visualizations were incorporated for explainability, showing the regions manipulated. Even with issues including data imbalance, high computational intensity, and generalization concerns, the system was found to be scalable and effective. Improvements for the future include incorporating audio detection, real-time deployment, and transformer-based architectures towards higher resilience