Deepfake detection using MobileNetV2 integrated with a support vector machine classifier

The rapid development of deep learning algorithms has led to a substantial improvement in the realism of synthetic media, particularly deepfakes, raising concerns about deception and digital security. Creating a lightweight deep learning architecture that can be trained on both fictitious and actual classes while achieving a respectable level of accuracy is the research contribution. The approach: This study suggests a deepfake identification technique that uses MobileNetV2, a lightweight convolutional neural network (CNN) for feature extraction, in conjunction with a Support Vector Machine (SVM) classifier as the final classifier. SVM takes advantage of MobileNetV2’s ability to rapidly extract high-level visual representations to produce a distinct separation between authentic and fraudulent content. The outcome: Experimental results demonstrate that the MobileNetV2-SVM system achieves competitive accuracy at a lower computational cost when compared to a conventional deep CNN classifier. These findings suggest that integrating MobileNetV2 and SVM enhances detection performance and offers a scalable solution for real-time processing and resource constraints. This approach contributes to the advancement of multimedia forensics and provides a useful and effective way to stop the spread of deepfake content. The evaluation findings showed an average training accuracy of 94.8%, precision of 93.5%, recall of 95.6%, and F1-score of 94.5%. 0.052 was the MSE value, 0.228 was the RMSE, and 0.041 was the MAE.