DEEPFAKES DETECTION USING DEEP LEARNING

ABSTRACT The rapid advancement of artificial intelligence and deep learning has enabled the creation of highly realistic synthetic media, popularly known as DeepFakes. While these technologies have legitimate applications in entertainment and education, they pose serious risks in misinformation, identity theft, and digital fraud. This study presents the design and implementation of a DeepFake Detection System using a hybrid Convolutional Neural Network–Long Short-Term Memory (CNN-LSTM) model to identify manipulated images and videos.Publicly available datasets—FaceForensics++, Celeb-DF v2, and the DeepFake Detection Challenge (DFDC)were used for training and evaluation. The system preprocesses input media through frame extraction, facial detection, normalization, and data augmentation to enhance learning. The CNN component extracts spatial features such as pixel inconsistencies and facial landmarks, while the LSTM network captures temporal dependencies like unnatural blinking and lip-sync irregularities.Experimental results demonstrated high performance across datasets, achieving 94.8% accuracy on FaceForensics++, 91.5% on Celeb-DF, and 87.3% on DFDC, confirming the model’s robustness and generalization ability. The system provides an intuitive user interface that allows media upload, real/fake classification, confidence scoring, and report generation.In conclusion, the hybrid CNN-LSTM model proved effective for DeepFake detection by leveraging both spatial and temporal features. The study contributes to digital media forensics by offering a scalable, accurate, and user-friendly solution capable of mitigating the growing threat of synthetic media manipulation. Future work should focus on enhancing real-time performance, multimodal analysis, and adaptive learning for evolving DeepFake techniques.