Deepfake Classification Project

Overview

This project focuses on classifying deepfake media using a machine learning model. The objective is to analyze a dataset containing metadata about various media files and predict whether they are real or fake using a Random Forest Classifier.

Dataset

The dataset deepfake_detection_metadata_dataset.csv contains 1000 rows of media metadata. It includes the following features:

• media_type: Image or Video
• content_category: News, Social Media, Interview, Political Speech
• face_count: Number of faces detected in the media
• audio_present: Whether audio is present
• lip_sync_score: Assessment of the lip sync quality
• visual_artifacts_score: Score indicating the presence of visual artifacts
• compression_level: Level of compression applied to the media
• lighting_inconsistency_score: Score evaluating lighting inconsistencies
• source_platform: Social media or news platform where the media was sourced
• label: Real or Fake

Methodology

The analysis runs in a Jupyter Notebook (deepfake_analysis.ipynb) and covers the following steps:

1. Data Loading and Exploration: Loading the data using pandas.
2. Data Cleaning: Dropping irrelevant columns.
3. Encoding: Converting categorical data into numeric values (One-Hot Encoding) and mapping the target label (Real to 0, Fake to 1).
4. Feature Scaling: Preprocessing numerical features using StandardScaler.
5. Model Training: Splitting the data into training (80%) and testing (20%) sets, then training a RandomForestClassifier.
6. Evaluation: The model is evaluated on the test set. An initial baseline using all features performs exceptionally well, while a more realistic evaluation excluding direct predictive artifacts yields performance closer to baseline, demonstrating the challenges in deepfake detection.

Requirements

• Python 3
• pandas
• scikit-learn
• Jupyter Notebook

Running the Project

1. Activate the provided virtual environment (venv).
2. Ensure required packages are installed (e.g., pip install pandas scikit-learn).
3. Start a Jupyter server and open deepfake_analysis.ipynb.
4. Run the notebook cells sequentially to reproduce the workflow.