Skip to content
Conference Open access

AI-Powered Zero Trust Network Security System on Cloud: A Hybrid Isolation Forest–Random Forest Detection Engine

Sep 2026 · 2026 International Conference on Secure IoT and Cybersecurity (ICOSICS) · pp. 1143-1148 · 0 citations · 20 references

Abstract

Perimeter-based defenses assume that traffic inside the network boundary is inherently safe, an assumption that breaks down in elastic, ephemeral cloud environments where a single compromised credential can enable lateral movement across an entire workload. This paper presents Phase 1 of an AI-driven Zero Trust Network Security System on Cloud that continuously inspects flowlevel traffic instead of trusting network location. An AWS-based pipeline, comprising an EC2 instance and a custom Virtual Private Cloud, captures live traffic and extracts a 15-feature flow-level schema, which, together with the NSL-KDD benchmark dataset, trains a hybrid dual-model detection engine (the CICIDS2017 benchmark is reserved for Phase 2 evaluation): an unsupervised Isolation Forest for novel-anomaly detection and a supervised, hyperparameter-tuned Random Forest for multi-class attack classification. On the held-out NSL-KDD test set, the Isolation Forest attains 83.32% binary detection accuracy (94.95% recall, AUC 0.939), while the Random Forest attains 75.52% five-class accuracy, both below the 90% target set for this phase; Section V analyzes this shortfall, attributing it mainly to minority-class Brute Force and Data Exfiltration attacks that closely resemble normal traffic at the flow level. The results establish a reproducible, cloudnative detection baseline that directly motivates the policy-enforcement layer planned for Phase 2.

Read PDF

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.