This study presents a novel approach to deepfake detection by integrating the DFRWS (Digital Forensics Research Workshop) framework with a deep learning architecture based on XceptionNet. The rapid advancement of deepfake technology poses a significant threat to digital media authenticity, necessitating robust and reliable detection methods. In this work, we implement a fine-tuned XceptionNet model enhanced with additional regularization techniques, specifically focusing on facial feature analysis. The model is trained on a balanced dataset comprising 2,000 images, equally divided between authentic and deepfake samples. Experimental results demonstrate exceptional performance, achieving an accuracy of 91.25%, precision of 88.73%, recall of 94.50%, and an AUC score of 0.9710. The proposed model shows a significant improvement in detecting subtle manipulation artifacts while maintaining computational efficiency, offering a promising solution for practical deepfake identification in real-world scenarios.