Optimizing CNN-based Deepfake detection with firefly algorithm: A Hybrid Approach

Abstract

Deepfake identification is crucial for ensuring the validity of digital material. This research seeks to improve the accuracy and efficacy of deepfake detection by improving a Convolutional Neural Network (CNN) using the Firefly Algorithm. We tested and compared the Firefly-optimized CNN model to a baseline CNN and the MobileNetV2 model on a dataset of authentic and fraudulent pictures. The CNN model was fine-tuned using the Firefly Algorithm, and MobileNetV2 served as a baseline for a more complicated design. The findings show that the Firefly-optimized CNN beats the baseline CNN and MobileNetV2 in classification accuracy, precision, recall, and ROC curve value. The Firefly-optimized CNN attained an accuracy of 99.09% and a ROC curve value of 0.99, outperforming the baseline CNN's accuracy of 95.98% and ROC value of 0.96, as well as MobileNetV2's accuracy of 93.48% and ROC value of 0.93. This research finds that adding optimization approaches such as the Firefly Algorithm may considerably improve CNN models' performance for deepfake detection, providing a solid answer for enhancing digital content verification procedures.