Natural Language Processing of HTTP Content for Insider Threat Detection Optimized by Modified Metaheuristic

This study explores the integration of Natural Language Processing (NLP) with the XGBoost classifier to establish a robust system for detecting insider threats, specifically focusing on the analysis of HTTP content within the domain of organizational cybersecurity. The study utilizes the TF-IDF technique to encode HTTP content extracted from an openly available insider threat dataset. To enhance performance, a thorough comparative analysis of contemporary optimization metrics is conducted to identify the most effective approach for the NP-hard task of hyperparameter selection. Introducing a customized version of the recently proposed sinh cosh optimizer (SCHO) algorithm, tailored to the unique demands of this research, significantly contributes to the methodology. The resulting model attains an impressive accuracy exceeding 97.6213%, showcasing substantial potential for real-world applications in the field of cybersecurity. This study not only addresses current limitations but also lays the groundwork for future advancements in insider threat detection and extends the proposed algorithm’s applicability to various pressing challenges in the cybersecurity landscape.