Deep neural network and model-based clustering technique for forensic electronic mail author attribution.

Area of Science:

• Computer Science
• Digital Forensics
• Cybersecurity

Background:

• Electronic mail is a major vector for cyber scams.
• Accurate author identification in emails is vital for digital forensics.
• Stylometry plays a key role in enhancing author attribution accuracy.

Purpose of the Study:

• To present a novel model for email author identification.
• To utilize deep neural networks and model-based clustering for attribution.
• To evaluate the model's performance on a large, imbalanced dataset.

Main Methods:

• Implementation of deep neural networks for email author attribution.
• Application of model-based clustering techniques for author identification.
• Experimentation on the publicly available Enron email dataset.

Main Results:

• Deep Neural Network (DNN) achieved 94% accuracy for 5 authors, 90% for 10, 86% for 25, and 75% for the entire dataset.
• The cluster-based technique achieved 86% accuracy on the entire dataset.
• High accuracy was maintained even on a highly imbalanced dataset.

Conclusions:

• The proposed models demonstrate high effectiveness in email author identification.
• These techniques offer robust solutions for digital forensic investigations.
• Stylometric analysis combined with advanced machine learning improves attribution reliability.