Insider Threat Identification From Accessed Website Content Optimized by Modified Metaheuristic

In today’s tech-driven landscape, safeguarding organizational security is imperative. Threats can arise externally or internally, necessitating preemptive measures. Despite the prevalence of insider risks like data breaches and intellectual property theft, literature lacks exploration of artificial intelligence (AI) potential in detecting malicious insiders. This study proposes leveraging natural language processing (NLP) and powerful classifiers to analyze employee behavior, offering an efficient means to mitigate insider threats. Effective classifiers depend on precise parameter selection, requiring optimization to align algorithms with specific problems. However, hyperparameter tuning faces NP-hard challenges due to extensive search spaces. Specialized algorithms like metaheuristic optimizers are commonly employed but require careful pairing with problems. In order to identify insider risks, this study presents a unique method that combines extreme gradient boosting (XGBoost) with bidirectional encoder representations using transformers (BERT) to analyze personnel HTTP content access. Leveraging NLP enhances robustness against common firewall evasion tactics. Additionally, a modified reptile search algorithm (RSA) algorithm is proposed for hyperparameter selection, achieving accuracy exceeding 97% on simulated security datasets. This framework offers a promising avenue for preempting insider threats and bolstering organizational security.