Systematic Review on Detection of Deepfake Attack using Machine Learning

The modern world has been dominated by the use of technology which determines the way how the activities are executed. Artificial intelligence (AI), during the last century, has been the center of all attention by permitting the execution of tasks that were unthinkable recently. Along with this, subsets of AI such as Deep Learning have been responsible for the occurrence of a type of attack that may cause dangerous consequences, the Deepfake attack. Deepfake attacks intend to manipulate the image, video, or audio of a person to produce information that was not created by the legitimate owner of the data. Countermeasures have been proposed, from traditional to Machine Learning. This last proves to be more suitable in the detection of the occurrence of the attack by the employment of differentiated algorithms. Has been employed a comprehensive analysis of principal algorithms used for Deepfake detection as methodology, highlighting their effectiveness in identifying manipulated media. The study revealed that ensemble models, especially those incorporating Convolutional Neural Networks (CNNs) as the base algorithm, persistently outperformed other approaches due the capability of this to capture complex spatial features in visual data, which has been defined as critical for detecting subtle anomalies introduced by Deepfake techniques. The review highlights strength of ensemble models, which can be gained by combining the strengths of multiple algorithms, enhancing detection accuracy and robustness, making them highly suitable for real-world applications.