Deepfake Detection Using XceptionNet with Advanced Frame-Level Video Analysis

Abstract

Deepfake systems have become increasingly capable due to advances in GANs, autoencoders, neural rendering, and AI-driven texture synthesis. This paper presents an XceptionNet-based deepfake detection model enhanced with frame-level preprocessing, dataset-driven optimization, architectural refinement, and interpretability components. A complete workflow overview, dataset behavior analysis, sample outputs, and adversarial vulnerabilities are included through figures. Additional technical details were incorporated from the project’s functional system, including workflow execution, layered architecture, and model explanation generation. This extended version provides deeper insight into implementation and realistic forensic application.