An Intelligent Risk Taxonomy for Enterprise Cybersecurity Using Advanced Deep Learning Techniques

Abstract

A sharp rise in sophisticated cyber intrusions, coupled with the well-known shortcomings of conventional defensive techniques, has pushed organizations to adopt smarter and self-adjusting strategies in their security operations. This study puts forward an intelligent risk taxonomy framework built on advanced deep learning methods to support precise threat identification across enterprise environments. Input records are sourced from the publicly available IDS Intrusion Dataset and then routed through a preprocessing module that handles missing entries and cleanses the raw values. A refined version of the Zebra–Gooseneck Barnacle Optimizer, here designated EZ-GBO, is applied to isolate ten informative attributes from the cleaned data. The reduced feature set is then passed into a Scalable and Adaptive Graph Neural Network (SAGNN), which both forecasts the risk category and labels every record under one of seven classes: DoS, DDoS, Brute Force, SQL Injection, Infiltration, Benign, and Bot. The entire pipeline is realized in Python and benchmarked against several competing techniques. Experimental outcomes show that the proposed scheme reaches an accuracy of 98.3 percent while keeping error magnitudes notably low.