Investigating the Use of Generative Adversarial Networks for Cybersecurity Dataset Training Data Substitution

Generative adversarial networks (GAN) in cybersecurity, particularly in natural language processing (NLP), remain underexplored. This study investigates the performance of machine learning (ML) techniques, specifically focusing on predicting spam emails. ML solutions offer customization and problem-specific tuning, making them well-suited for this task. The applied NLP approach employs term frequency-inverse document frequency (TF-IDF) for text mining, evaluating term importance based on overall appearance across documents and frequency in the evaluated document. The framework proves crucial for detecting patterns, especially in the absence of unfiltered spam email data. To address data scarcity, the study evaluates ML predictions on both real and GAN-generated data. The comparison reveals that the best models, trained on real and synthetic data, achieve over 97% accuracy, suggesting GANs as a promising approach to overcome limitations associated with limited secure data.