FKAN-Enhanced Multi-View Self-Supervised Network Intrusion Detection with Frequency-Aware Temporal Modeling

Overview

This project introduces a contrastive multi-view self-supervised framework for network intrusion detection (NID) that integrates finite-difference temporal modeling with Fourier Kolmogorov–Arnold Network (FKAN)–enhanced temporal encoding. Criteria: (1) Accuracy - Should be able to identify a high number of attacks while maintaining a low false positive rate. (2) Efficiency - Should be able to handle large volumes of multi-feature data in a short amount of time. (3) Versatility - Should be able to detect various types of intrusions, especially zero-day attacks without known signatures. The method processes heterogeneous traffic representations—including raw packets, flow statistics, and finite-difference sequences—using transformer-based encoder-decoder modules. A cross-feature correlation view enhances the modeling of inter-feature relationships, while FKAN-augmented temporal embeddings strengthen the modeling of periodic and long-range temporal behaviors. A contrastive objective aligns each reconstruction with its original view and enforces diversity across reconstructions, effectively combining generative and discriminative principles. Once trained, the framework embeds traffic into a shared latent space and detects anomalies as deviations from the benign cluster, particularly those exhibiting irregular temporal patterns. Experiments on several real-world datasets show consistent improvements over state-of-the-art unsupervised and self-supervised NID systems, demonstrating the approach’s effectiveness and scalability. On the CIC-IDS2017 network intrusion dataset, the initial model achieved an AUC-ROC score of 0.978, an AUC-PR score of 0.976, and an F1-score of 0.946. After implementing the FKAN and other adjustments, the AUC-ROC score improved to 0.996, while the F1-score improved to 0.965. Similar results and improvements were observed on UNSW-NB15, another well-known network intrusion dataset.

Competition history

• CSEF 2026 Computational Science (Senior Division) · Entry S-07-34

Related projects

Closest projects by meaning, across every fair and year in the corpus.

Browse more like this

• All Computational Sciences projects
• CSEF Computational Sciences projects
• Every CSEF 2026 project