Published Paper Details:

Deepfake technology uses advanced artificial intelligence and deep learning techniques to generate highly realistic manipulated videos that imitate human appearance, facial expressions, and behavior. The rapid growth of deepfake generation methods has created significant challenges related to misinformation, identity fraud, cybersecurity, and digital media authenticity. Detecting manipulated content has become increasingly difficult due to continuous improvements in visual quality and synthesis techniques. This study presents an effective deepfake video detection framework based on EfficientNet-B0 and Vision Transformer architectures. EfficientNet-B0 serves as the primary feature extraction model, capturing subtle facial artifacts such as texture inconsistencies, blending errors, and unnatural boundaries, while the Vision Transformer analyzes global contextual relationships through an attention mechanism. A structured preprocessing pipeline involving frame extraction, face detection, cropping, normalization, and data augmentation is employed to enhance model robustness and generalization. The framework is trained and evaluated using benchmark deepfake datasets containing diverse manipulation techniques. A Flask based web application is integrated for video upload and prediction. Experimental results demonstrate high detection accuracy, reliability, and effectiveness in distinguishing real and manipulated videos.