Mitigating Server-Side Request Forgery (SSRF) Attacks: An Empirical Analysis of Deep Learning-Based Approaches

Amid the escalating wave of cybersecurity threats, server-side request forgery (SSRF) has emerged as a critical concern, presenting significant risks to organizations. This paper undertakes a comprehensive empirical analysis of deep-learningbased mitigation techniques specifically designed to counteract SSRF attacks. The evaluation focuses on two prominent deep learning models, Long Short-Term Memory (LSTM) and Bidirectional Encoder Representations from Transformers (BERT), utilizing the CIC dataset of URLs. Our study demonstrates that the LSTM model surpasses BERT, achieving an accuracy of 98.2%, compared to BERT's 94%. Notably, our LSTM accuracy of 98.2% outperforms the current literature's reported accuracy of 96%. This notable improvement underscores the efficacy of the LSTM model in effectively addressing SSRF challenges. It is imperative to highlight that the successful fine-tuning and training of deep learning models require substantial computational resources, a critical aspect underscored by our study. Future research directions include exploring alternative deep learning models, and hybrid approaches, and evaluating model efficacy across diverse network environments. This research contributes valuable insights to the SSRF mitigation domain, accentuating the significance of rigorous evaluation for effectively addressing real-world cybersecurity challenges.