Application of Automl Technologies for Automated Detection of Web Attacks in Network Traffi c Based on AutoGluon Tabular
Year: 2026
Type: article
Abstract: The increasing complexity and scale of network attacks necessitates the transition from traditional signature intrusion detection systems (IDS) to more adaptive approaches based on machine learning methods. This study presents a comparative analysis of the eff ectiveness of classical machine learning algorithms and the automated machine learning (AutoML) framework AutoGluon in solving the problem of multiclass classifi cation of network traffi c. The publicly available CICIDS2017 dataset was used as an experimental base, which includes both legitimate connections and various types of attacks that simulate real-world network conditions. An analysis of network characteristics and their discriminative ability is carried out, as well as a detailed assessment of model performance using the metrics of general accuracy, F1-measure, and training time. The results showed that ensemble algorithms, in particular Random Forest and AutoGluon, achieve the highest overall accuracy (more than 99.5%). At the same time, the detection effi ciency of minor classes of attacks, such as SQL Injection and cross-site scripting (XSS), is critically low. It has been established that this problem is caused not only by a strong data imbalance, but also by the characteristics of the attacks themselves, which are often disguised as legitimate traffi c and do not form pronounced statistical anomalies. Thus, the study demonstrates the potential and limitations of using AutoML frameworks in the fi eld of cybersecurity. The practical signifi cance of the work lies in the possibility of reducing the development time of intrusion detection systems (IDS) while maintaining high accuracy, and the prospects for further research are related to the integration of class balancing methods and the development of hybrid models to improve the recognition of rare attacks.
Authors: M. K. Arabov, A. S. Shumikhina
Institution: Kazan Federal University
Language: English
Subfield: Computer Networks and Communications
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7166881844