DeepFake Image Detection Using Convolutional Neural Network with EfficientNet Architecture

The growing sophistication of generative Artificial Intelligence (AI) has intensified the threat posed by deepfake technologies, which are capable of producing highly realistic yet fabricated facial images and videos. These manipulated visuals can mislead the public, infringe on personal privacy, and damage reputations. This study aims to develop an effective deepfake image detection system using Convolutional Neural Networks (CNN) enhanced with EfficientNet architectures (B3–B5). The research adopts the Cross Industry Standard Process for Data Mining (CRISP-DM) methodology, providing a structured data science framework that spans from problem definition to deployment. Three open-access datasets (Celeb-DF v2, DeeperForensics-1.0, and DFDC) are utilized to train and evaluate the models. Experimental results show that EfficientNet-B5 achieves the highest classification accuracy at 93.2%, outperforming both the baseline CNN and other EfficientNet variants. The proposed method demonstrates strong cross-dataset generalization and computational efficiency, making it suitable for real-world applications. This research contributes a comparative evaluation of scalable deepfake detection models, practical deployment insights, and a foundation for future work in explainable and real-time AI-based media forensics.