Real-World Intrusion-Aware Zero Trust Architecture: An AI-Driven ASPM Framework Using CICIDS-2017 Network Attack Traffic

The rise of enterprise networks and the general popularity of advanced cyber threats being targeted at organizations, specifically zero-day attacks that are able to circumvent traditional signature-based security systems. This paper presents an Intrusion-Aware Zero-Trust Modelling Framework with Application Security Posture Management (ASPM) to provide a suitable model for application security, using data from the CICIDS-2017 dataset. The framework achieves very accurate traffic classification using ANN, Random Forest, and XGBoost models. In addition to regular intrusion detection, the ANN model builds and provides a low-risk score from available input data, allowing organizations to make zero-trust access decisions and inform ASPM policies related to attack awareness. The experimental findings indicate that ANN outperforms other models, achieving 98.56% accuracy, 96.01% F1-score, and 0.998 AUC, whereas XGBoost offers a good trade-off between accuracy and real-time performance. The suggested models were also compared with baseline models, such as Logistic Regression, AdaBoost, and CNN-LSTM, which demonstrated worse detection performance. The proposed framework is thus a scalable, adaptive, and practical means of real-time intrusion detection and the enforcement of zero-trust policies in modern networks.