Fusion of Machine Learning and Deep Learning:A Hybrid Approach for Deepfake Detection
Aditya Helode, Akash Yadav, Vishnu Prasad Verma, Krishnarajanagar G. Srinivasa · 2024
Deepfakes refer to artificially generated digital media designed to produce highly realistic fake videos, often with the intention of deceiving the viewer. Techniques like Generative Adversarial Networks (GANs) are commonly employed to achieve this goal. These methods generate content that closely resembles real visuals, making it challenging for conventional detection techniques to differentiate between real and fake content. Detection systems often rely on Convolutional Neural Network (CNN) based discriminators to identify such synthesized media. However, these systems mainly focus on the spatial characteristics of individual video frames and may struggle to capture temporal information from the relationships between frames. This paper employs a hybrid methodology that combines machine learning and deep learning techniques to address the issue of deepfakes. Initially, we utilized a CNN model, specifically VGG16, for extracting features. Subsequently, we employed two machine learning models, Random Forest and XGBoost, to classify the data into real and fake categories. The Random Forest-CNN model achieved a higher accuracy compared to the XGBoost-CNN model, which in turn surpassed many other techniques in terms of accuracy.