Deepfake Detection Models Struggle in Real-World Applications

• •Detection models excel in lab settings but fail in real-world scenarios.
• •Significant challenges include cross-dataset generalization and adversarial robustness.
• •The review categorizes detection methods into eight distinct approaches.

A systematic review published in the International Journal of Data Science and Analytics highlights the limitations of deepfake detection models, which perform well in controlled lab settings but struggle with real-world applications. The review synthesizes findings from 101 studies conducted between 2021 and 2025, revealing that while many models achieve over 95% accuracy in ideal conditions, they face significant challenges in cross-dataset generalization and adversarial robustness. The authors categorize detection approaches into eight types, including spatial, temporal, and multimodal methods, each exploiting different weaknesses of synthetic media. Ethical and societal implications of deepfakes, such as misinformation and privacy violations, are also discussed. The review emphasizes the need for an integrated framework that addresses the technical, ethical, and legal dimensions of deepfake detection.