Cognitive Cyber Crimes in the Era of Artificial Intelligence

17Deepfake Forensics and Synthetic Media Attacks

Yagyanath Rimal^1*, Rajesh Kumar Chakrawarti², Neeraj Mehta³, and Alpesh Soni³

¹Department of Computer Science, Pokhara University, Pokhara, Nepal

²Sushila Devi Bansal College of Technology, Indore, Madhya Pradesh, India

³Department of CSE, SVIIT, SVVV, Indore, India

Abstract

The exponential rise in synthetic media, particularly deepfakes, poses severe threats to digital integrity, societal trust, and cybersecurity. This study proposes an advanced deepfake forensic framework using the Deepfake Detection Challenge dataset, comprising over 100,000 labeled videos with real and synthetic manipulations. A hybrid detection technique titled SPFN-RADNet (Spatial–Phase Fusion Network With Residual Attention Discriminator Network) is introduced, integrating spatial texture patterns and phase-aware representations to enhance detection precision. The proposed method achieved an accuracy of 98.76%, area under the curve of 0.984, F1 score of 0.973, and EER (equal error rate) of 1.89%. Results highlight its superior performance against conventional convolutional neural network and transformer-based models. The approach establishes a benchmark for real-time and robust detection of synthetic media threats.

Keywords: Deepfake forensics, synthetic media, SPFN-RADNet, DFDC, deep learning

17.1 Introduction

The proliferation of artificial intelligence (AI)–generated [1, 2] media, especially deepfakes, has revolutionized digital content creation while ...