International Journal of Scientific Research and Engineering Development

The system is developed using Python and integrates a Streamlit-based user interface for real-time interaction. The dataset consists of real and deepfake images and videos, which are preprocessed through face detection, frame extraction, and normalization techniques. A Convolutional Neural Network (CNN) model is trained on this data to classify media as real or fake with an observed accuracy of approximately 88–95% depending on dataset variations. Performance evaluation shows that the average prediction time per input is approximately 0.8 to 1.5 seconds, ensuring near real-time response for users. The system also incorporates Grad-CAM (Gradient-weighted Class Activation Mapping), enabling visualization of important regions influencing the model’s decision. Experimental results indicate that the proposed system effectively detects deepfake content and provides visual explanations for predictions. The combination of efficient execution time, satisfactory accuracy, and explainability makes the system a practical solution for digital media verification. The system can be further extended with real-time detection and advanced deep learning models for improved performance.

Shifa Bilal Tamboli, Simeen Phiroj Mulani, Arman Tajuddin Shiakh,"Explainable Deepfake Detection Using GRAD-CAM" International Journal of Scientific Research and Engineering Development, V9(2): Page(2363-2364) Mar-Apr 2026. ISSN: 2581-7175. www.ijsred.com. Published by Scientific and Academic Research Publishing.