A STUDY OF SGD AND ADAM APPROACHES TO TRAINING AN LSTM ARTIFICIAL NEURAL NETWORK FOR MALICIOUS TRAFFIC RECOGNITION
Alexandr I. Timoshenkov
Moscow Aviation Institute, Moscow, Russia, tim2_02@mail.ru
Anastasia Y. Kudryashova
Moscow Technical University of Communications and Informatics, Moscow, Russia, a.i.kudriashova@mtuci.ru
DOI: 10.36724/2664-066X-2025-11-4-9-14
SYNCHROINFO JOURNAL. Volume 11, Number 4 (2025). P. 9-14.
Abstract
This paper examines the problem of binary classification of network traffic using an artificial neural network (ANN) based on the LSTM (Long Short-Term Memory) architecture. A comparative study of the effectiveness of two popular optimizers – stochastic gradient de-scent (SGD) and adaptive moment estimation (Adam) – Is conducted on various network attack scenarios from the CICIDS-2017 dataset. The focus is on classification quality metrics: accuracy, recall, prediction accuracy, and F1-score. Experiments demonstrate that the Adam optimizer demonstrates higher and more stable performance, especially under conditions of significant class imbalance characteristic of real-world network traffic. A detailed theoretical justification for the advantages and disadvantages of each optimizer is provided, and the causes of the observed experimental phenomena are analyzed in detail.
Keywords: intrusion detection; anomaly detection; neural network; LSTM; SGD; Adam; binary classification; malicious traffic; CICIDS-2017
References
[1] B. B. Borisenko, S. D. Erokhin, A. S. Fadeev, I. D. Martishin, “Detection of computer attacks using a multilayer perceptron and long short-term memory networks,” Systems for synchronization, formation and processing of signals. 2021. Vol. 12, No. 5, pp. 4-13.
[2] S. S. Galizdra, A. Yu. Kudryashova, “Method of biometric identification of a person by a row of teeth based on a photograph with an open smile,” Systems for synchronization, formation and processing of signals. 2024. Vol. 15, No. 6, pp. 34-39.
[3] A. Yu. Kudryashova, A. A. Karavanova, “An encryption algorithm for hard drive partitions to protect against intruders,” Telecommunications and Information Technologies. 2024. Vol. 11, no. 2, pp. 32-37.
[4] www.unb.ca | Intrusion detection evaluation dataset (CIC-IDS2017) / [Electronic resource] // URL: https://www.unb.ca/cic/datasets/ids-2017.html
[5] studfile.net | Back Propagation Learning Algorithm (Back Propagation – bp)./ [Electronic resource] // URL: https://studfile.net/preview/21852300/page:6
[6] www. vc.ru | Optimizers (Adam, SGD) /[Electronic resource] // URL: https://vc.ru/id4616024/2263731-optimizatory-adam-i-sgd-upravlenie-shagami-obucheniya-nevrosotey
[7] education.yandex | 15.4. Optimization Methods in Deep Learning /[Electronic resource] // URL: https://education.yandex.ru/handbook/ml/article/metody-optimizacii-v-deep-learning/
[8] cyberleninka.ru | Synthesis of a Machine Learning Model for Detecting Computer Attacks Based on the CICIDS-2017 Dataset/[Electronic resource] // URL: https://cyberleninka.ru/article/n/sintez-modeli-mashinnogo-obucheniya-dlya-obnaruzheniya-kompyuternyh-atak-na-osnove-nabora-dannyh-cicids2017
[9] K. O. Safronov, A. Yu. Kudryashova, Yu. V. Molodtsova, “Study of the Relationship between AI Hallucinations, Prompt Length, and Logical Paradoxes: The Role of Kolmogorov Complexity and Semantic Analysis in Ensuring the Integrity of Information Systems,” REDS: Telecommunication Devices and Systems. 2025. Vol. 15, No. 3, pp. 22-26.
[10] S. S. Galizdra, A. Yu. Kudryashova, “Method of biometric identification of a person by a row of teeth based on a photograph with an open smile,” Systems for synchronization, formation and processing of signals. 2024. Vol. 15, No. 6, pp. 34-39.
[11] A. Y. Kudriashova, S. S. Galizdra and N. V. Toutova, “Designing Implementation of Additional Modules to Increase the Security and Stability of the Biometric Identification System,” 2025 Wave Electronics and its Application in Information and Telecommunication Systems (WECONF), St. Petersburg, Russian Federation, 2025, pp. 1-5, doi: 10.1109/WECONF65186.2025.11017218.
[12] N. V. Toutova, A. Y. Kudriashova, and S. S. Galizdra, “Implementation of Additional Modules to Increase the Security and Stability of the Biometric Identification System,” 2025 Wave Electronics and its Application in Information and Telecommunication Systems (WECONF), St. Petersburg, Russian Federation, 2025, pp. 1-5, doi: 10.1109/WECONF65186.2025.11017156.
[13] A. Yu. Kudryashova, V. A. Zakharova, “Development of information security measures for defense industry enterprises to implement the Digital Economy 2030 policy,” Telecommunications and Information Technologies. 2024. Vol. 11, No. 2, pp. 45-51.
[14] A.Yu. Kudryashova, “Development of a program for calculating additional distortions for various models of errors”, T-Comm, 2022. vol. 16, no.1, pp. 51-58. DOI: 10.36724/2072-8735-2022-16-1-51-58