Theses and Dissertations

Advisor

Gudla, Charan

Committee Member

Trawick, George

Committee Member

Chen, Jingdao

Date of Degree

8-12-2026

Original embargo terms

Visible MSU Only 6 months

Document Type

Graduate Thesis - Campus Access Only

Major

Cybersecurity & Operations

Degree Name

Master of Science (M.S.)

College

James Worth Bagley College of Engineering

Department

Department of Computer Science and Engineering

Abstract

Modern network intrusion detection systems (NIDS) suffer f rom a n umber o f fundamental architectural limitations. One limitation is that NIDS classify flows only after the flow has fully terminated which introduces detection delays leaving systems vulnerable to ongoing attacks. The second architectural limitation is that most NIDS network traffic as in dependent ta bular rows which ignores rich relational structure that exists between communications hosts. To address these limitations, we introduce a model called MemFlow-IDS which is a novel framework that reformulates network intrusion detection as link-level classification over a continuous-time dynamic graph (CTDG). We specifically apply a memory-augmented temporal graph learning network to these problems. We evaluate MemFlow-IDS on the CICIDS-2017 dataset on two task dimensions which are early evaluation and inductive testing. The model performs excellent for both tasks in F1-scores, precision, and recall. The research demonstrates that CTDG modeling provides a stable model for intercepting cyber threats among real-world enterprise networks.

Download

Share

COinS