Fake Person Detection Using Machine Learning

Welcome to our project repository for the Fake Person Detection using Machine Learning, developed as part of our coursework at the University of Applied Science, Potsdam. This project focuses on utilizing advanced neural network architectures, specifically CNN and ResNet, to accurately identify fake videos from real ones based on image data.

Project Overview

The advent of sophisticated generative modeling has made the creation of realistic fake identities more accessible, posing serious threats to digital security and privacy. Our project aims to address these challenges by developing effective machine learning tools to distinguish real identities from fake ones.

Key Features

• Two Machine Learning Models: Implementation and training of two different models, CNN and ResNet, to evaluate their effectiveness in fake person detection.
• Kaggle Dataset: Utilized a robust dataset from Kaggle, facilitating structured learning and model development.
• Performance Analysis: Detailed analysis of model accuracy and loss to refine and optimize performance.

Getting Started

Prerequisites

Ensure you have Python 3.8+ installed on your system. You can download it from Python's official site.

Technologies Used

• Python: Main programming language used for the project.
• TensorFlow and Keras: For building and training the neural network models.
• NumPy and Pandas: For data manipulation and analysis.

Results

Our comparative analysis shows the robustness and accuracy of the CNN model over the ResNet model in detecting fake identities with a higher level of generalizability.

Contributions

• Ivan Logutov: Led the model development and dataset structuring.
• Artur Yurchenko: Assisted in model training and performance evaluation.
• Vitaliy Danyuk: Focused on writing the project report and preparing the presentation.

Future Directions

We aim to refine these models further by integrating state-of-the-art tools for deployment on websites and applications, ensuring practical utility and enhancing digital trust.

License

This project is licensed under the MIT License - see the LICENSE.md file for details.

Acknowledgments

• Thanks to Kaggle for providing a comprehensive dataset that has been instrumental in our training processes.
• Gratitude to our mentors at the University of Applied Science, Potsdam, for their guidance and support.