Abstract

Employing AI and the Isolation Forest algorithm, this study presents a novel method for predicting and detecting anomalies in the Core Backbone Flow (CBF) traffic of secure internet architectures. In a highly controlled environment, the unsupervised machine learning approach to CBF traffic aims to discover patterns associated with potential security breaches or system performance issues. The CBF traffic patterns were collected from controlled experimental networks, enabling the Isolation Forest model to be trained and tested for outlier detection. For real-time use cases, the system demonstrated high accuracy and a low false-positive rate, surpassing traditional benchmarks of precision, recall, F1 score, and ROC-AUC. Assessing the model alongside traditional outlier detection methods highlighted its novel capability to detect unseen anomalies without the use of labeled datasets, thereby showcasing a clear advantage over other methods. AI-enhanced forecasting capabilities further bolster the model's defenses by enabling anticipatory action to neutralize threats before they significantly escalate, thereby fortifying predictive paradigms in cybersecurity. This initiative supports the shift in cybersecurity paradigms toward proactive rather than reactive measures by enabling automated systems to counter evolving threats and dynamically disengage from signature-based measures. The model's utility in this context is straightforward, but its potential relevance to large, heterogeneous networks remains underexplored. The internet framework's security is significantly enhanced with these new technologies, focusing on addressing modern adaptive threats.