Zero Trust Architecture Dataset (ZTAD): Advancing AI-Driven Cybersecurity Through Multi-Source Behavioral Analysis and Feature Optimization

Cyberattacks are continuously becoming more sophisticated, especially with the proliferation of Living off the Land Binaries, which bypass traditional signature-based defenses. To meet this challenge, this research paper enhances a methodology for creating a comprehensive cybersecurity dataset aligned with the NIST Zero Trust Architecture (ZTA) and labeled using the MITRE ATT&CK framework. The proposed Zero Trust Architecture Dataset (ZTAD) captures diverse attack vectors and behaviors to support robust, AI-driven threat detection and prediction. The work aims to develop a multi-source, multi-asset dataset reflective of real-world environments, and to implement a proof-of-concept leveraging open-source tools to emulate a ZTA testbed. In order to enhance Artificial Neural Network (ANN) detection capabilities initially with an accuracy of 99,6%, this work optimizes the feature selection using Evolutionary Algorithms (EA). Experimental results demonstrate the efficiency of the ZTAD and the related EA-ANN hybrid model with an accuracy of 99,9%. This work addresses the critical gaps in the state-of-the-art cybersecurity datasets and methodologies, enabling advanced behavioral analysis and proactive security.