Original scientific paper

https://doi.org/10.17559/TV-20250827002928

Enhancing Machine Learning for Anomaly Detection and Classification Using Entropy-Based Dataset Enrichment

Igor Fosić ; HEP-Telekomunikacije d.o.o., M. Divalta 199, HR-31000 Osijek *
Drago Žagar ; Josip Juraj Strossmayer University of Osijek, Faculty of Electrical Engineering, Computer Science and Information Technology Osijek, Kneza Trpimira 2B, HR-31000 Osijek

* Corresponding author.

Abstract

In machine learning and classification, entropy holds significant potential. This paper introduces a method to calculate Shannon entropy across all features within individual records in four IDS datasets: CSE-CIC-IDS2018, CIC-IDS2017, UNSW-NB15, and LUFlow. Each dataset is reshaped according to NetFlow, allowing for easy acquisition from actual network devices. A comparison of dataset versions with and without the entropy feature showed that the proposed entropy calculation method improves classification performance, even though the number of features was reduced compared to the original dataset. Enhanced classification and anomaly detection results are evident through improved AUC metrics and confusion matrix outcomes.

Keywords

anomaly detection; dataset enrichment; entropy; machine learning; NetFlow

Hrčak ID:

348690

URI

https://hrcak.srce.hr/348690

Publication date:

30.6.2026.

Visits: 0 *