A Hybrid Approach to Spam Detection Using UNet and Diffusion Models

Ziyan Li

2024

Abstract

As the threat of spam emails continues to rise, effective classification is critical for ensuring email security, particularly in countering phishing and malware attacks. This study introduces a hybrid approach that combines the U-shaped Network (UNet) model and the Diffusion Model to enhance spam detection accuracy. The research utilizes a balanced dataset of legitimate and spam emails, leveraging the strengths of both models. The Unet model, with its encoder-decoder architecture, achieved a training accuracy of 90% and a validation accuracy of 80% after 50 epochs, demonstrating strong feature extraction capabilities. In contrast, the Diffusion Model, designed to handle noisy and obfuscated data, achieved a training accuracy of 88% and a validation accuracy of 75%. Although the UNet model excelled in general classification tasks, the Diffusion Model proved more effective in handling complex and disguised spam patterns. The experimental results suggest that combining these models could further improve spam detection across diverse scenarios. Future work will focus on optimizing the system for real-time spam detection and enhancing its ability to generalize across various types of spam emails.

Paper Citation

in Harvard Style

Li Z. (2024). A Hybrid Approach to Spam Detection Using UNet and Diffusion Models. In Proceedings of the 2nd International Conference on Data Analysis and Machine Learning - Volume 1: DAML; ISBN 978-989-758-754-2, SciTePress, pages 359-364. DOI: 10.5220/0013517700004619

in Bibtex Style

@conference{daml24,
author={Ziyan Li},
title={A Hybrid Approach to Spam Detection Using UNet and Diffusion Models},
booktitle={Proceedings of the 2nd International Conference on Data Analysis and Machine Learning - Volume 1: DAML},
year={2024},
pages={359-364},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0013517700004619},
isbn={978-989-758-754-2},
}

in EndNote Style

TY - CONF

JO - Proceedings of the 2nd International Conference on Data Analysis and Machine Learning - Volume 1: DAML
TI - A Hybrid Approach to Spam Detection Using UNet and Diffusion Models
SN - 978-989-758-754-2
AU - Li Z.
PY - 2024
SP - 359
EP - 364
DO - 10.5220/0013517700004619
PB - SciTePress