Leveraging Metaheuristic Optimization to Enhance Insider Threat Detection Through Email Content Natural Language Processing

Ensuring operational integrity in the ever-evolving internet landscape requires robust cybersecurity measures to combat emerging security compromise vectors. Despite traditional techniques, insider threats pose a significant challenge, exploiting the human element and contributing to substantial cyberattack losses. On-premises attacks, facilitated by physical access, defy conventional detection methods, necessitating alternative approaches. This study explores the integration of natural language processing (NLP) with the AdaBoost classifier for identifying malicious intent in email content, utilizing publicly available datasets to detect insider threats. Optimizing model hyperparameters through various algorithms, including a modified elk herd optimization algorithm, enhances performance. The best-performing model is analyzed in-depth, uncovering key terms crucial for malicious intent detection and their roles in the classification process.