Comparitive Literature Review on Deepfake Detection Techniques

Deepfake technology has emerged as a significant challenge to the authenticity and trustworthiness of digital media. This paper presents a comparative literature review of eight leading deepfake detection methods, including Spatiotemporal Convolutional Networks, PRNU-based Detection, Frequency Domain Filtered Residual Networks, Deep Learning-based Approaches, Transfer Learning models, Multimodal Detection techniques, and Generative Adversarial Network (GAN)-based frameworks. Each method is evaluated based on accuracy, robustness, computational efficiency, and adaptability across benchmark datasets such as FaceForensics++, Celeb-DF, and UADFV.The analysis identifies the Frequency Domain Filtered Residual Network as the most reliable framework, achieving over 98% accuracy and strong generalization on uncompressed data. Ethical and societal implications — including privacy, fairness, and transparency — are discussed to promote responsible AI research. The paper concludes by outlining strategies for improvement and future directions, including real-time detection, multimodal fusion, explainable AI, and privacy-preserving architectures. This comparative review serves as a reference point for advancing the reliability and accountability of deepfake detection systems in modern digital forensics.