Cyber Threat Detection Using an Integrated AI System

Abstract

This work presents the design and implementation of Cyber Threat Detection Using an Integrated AI System, an artificial intelligence–based framework developed for proactive cyber threat detection and analysis. The proposed platform combines network intrusion detection and phishing risk assessment into a unified architecture to address modern multi-vector cyberattacks. Benchmark datasets such as CICIDS-2017 for network traffic and URL-based datasets including PhishTank and UCI Repository will be utilized for model training and evaluation. The system employs an Unsupervised Deep Learning Autoencoder to learn normal network behavior and detect zero-day intrusions through reconstruction error analysis. In parallel, a Random Forest classifier analyzes lexical URL features such as domain entropy, special character frequency, URL length, and HTTPS validity to identify phishing websites in real time. The architecture follows a modular microservices design integrated with a FastAPI backend and an interactive dashboard interface. Additionally, Generative AI–based alert summarization provides human-readable threat explanations to reduce analyst fatigue and improve response efficiency. The expected outcome is enhanced detection of zero-day attacks and phishing threats, reduced false positives, improved situational awareness, and a scalable, intelligent cybersecurity solution capable of adapting to evolving digital threats while supporting faster and more informed security decision-making.