ML Based Intrusion Detection System using NLP

As cyberattacks become increasingly sophisticated, web-based applications remain prime targets for malicious actors. These attacks, such as SQL Injection, Cross-Site Scripting, Buffer Overflow, etc., exploit vulnerabilities in web servers through crafted HTTP requests. To mitigate these threats, we present a Machine Learning-based Intrusion Detection System (IDS). This IDS uses models that are trained on vector embeddings of HTTP requests. These vector embeddings are generated using NLP transformers such as Gensim, USE, and Sentence-BERT after preprocessing the HTTP requests. Among different ML models, the Support Vector Machine classifier trained on sBERT embeddings performed best with 99.905% accuracy.