Automated Detection of Deepfakes using Integrated AI and Computer Vision Strategies

Title

Automated Detection of Deepfakes using Integrated AI and Computer Vision Strategies

Creator

Varkey, Tojin; George, Jossy; Chanti, S.

Description

Deepfakes, or artificial intelligence-generated fake videos, are becoming a greater concern for online information trust, personal privacy, and digital content security. This paper presents a straightforward and understandable technique for automatically identifying deepfakes in order to address this significant problem. The approach makes use of conventional computer vision and machine learning methods. The model examines manually produced visual cues such as eye distance, mouth movement, and head tilt in video footage. To increase accuracy, it employs a variety of classifier types, including Random Forest, Gradient Boosting, and a soft Voting Classifier. A method known as SMOTE was used to clean and balance the data, and categorical data was transformed into a format suitable for machine learning models. With an F1-score of 0.9802 and 98% accuracy, the results demonstrate that the Voting Classifier, which combines several models, works admirably while being straightforward and effective. This method makes detection successful and simple to comprehend while offering a helpful tool for swiftly identifying deepfakes, especially on systems with constrained resources. 2025 IEEE.

Source

2025 World Skills Conference on Universal Data Analytics and Sciences, WorldSUAS 2025;

Date

01-01-2025

Publisher

Institute of Electrical and Electronics Engineers Inc.

Subject

Computer Vision; Deepfake Detections; Ensemble Learning; Gradient Boosting; Random Forest; SMOTE

Coverage

Varkey T., CHRIST (Deemed to Be University), India; George J., CHRIST (Deemed to Be University), India; Chanti S., CHRIST (Deemed to Be University), India

Rights

Restricted Access; Hardcopy may be available in the library

Relation

ISBN: 979-833153925-2;

Format

online

Language

English

Type

Conference paper

Identifier

https://doi.org/10.1109/WorldSUAS66815.2025.11199153

https://www.scopus.com/pages/publications/105022291507?origin=resultslist