View Proposal

---

Proposer

Inkar Zhumay

Title

Machine Learning for Network Intrusion Detection

Goal

Description

For Aktobe Students Only. Train several intrusion detection models on a standard labelled traffic dataset and establish their baseline detection performance. Construct evasion attacks that perturb flow features within constraints that keep the underlying attack operational — padding, timing jitter, fragmentation and traffic shaping are realisable, while arbitrary changes to derived features such as flow duration or byte counts are not. This distinction is the core of the project: unconstrained feature-space attacks overstate vulnerability and are frequently reported in the literature without qualification. Measure detection rate under increasing perturbation budget, and evaluate whether adversarial training, feature-set restriction or ensembling recovers robustness and at what cost to clean-traffic false alarm rate.

Resources

Background

Url

Difficulty Level

Moderate

Ethical Approval

None

Number Of Students

1

Supervisor

Inkar Zhumay

Keywords

machine learning, network security, intrusion detection, cybersecurity, network traffic analysis

Degrees

Bachelor of Science in Computational Sciences and Software Engineering