Improving a machine learning method for an automated control system ⋆ Viktoriia Zhebka1,†, Oleksii Ananchenko1,†, Kateryna Osadcha2,3,*,†, Serhii Zhebka1,† and Andrii Aronov1,† 1 State University of Information and Communication Technologies, 7 Solomenskaya str., 03110 Kyiv, Ukraine 2 Norwegian University of Science and Technology, 1 Høgskoleringen, 7034 Trondheim, Norway 3 Bogdan Khmelnitsky Melitopol State Pedagogical University, 59 Naukove mistechko str., 69000 Zaporizhzhia, Ukraine Abstract The paper is devoted to the improvement of the automated learning management system by integrating the metric proximal gradient method. Improving the automated learning management system helps to increase the efficiency, quality, and safety of the educational process by automating routine tasks and implementing individualized curricula. The paper discusses the use of machine learning to analyze student performance and detect suspicious user activity, which increases the transparency and reliability of the system. The use of the metric proximal gradient method ensures efficient solutions to optimization problems and increases the adaptability of the model in a dynamic educational environment. Also the paper presents an improved approach to automated learning management systems through the implementation of an advanced machine learning method based on the metric proximal gradient algorithm. The research addresses key challenges in educational process management, including system efficiency, quality assurance, and security enhancement. The proposed method incorporates a specialized database for comprehensive event logging and implements clustering and regression algorithms for student performance analysis. The improved metric proximal gradient algorithm demonstrates effective convergence properties through diagonal step sizing and non-monotonic linear search strategies. Results indicate that this approach provides enhanced optimization capabilities for handling complex data structures and adapting to dynamic educational environments. The implementation shows particular promise in personalizing educational routes, optimizing curricula, and maintaining system security through automated anomaly detection. Keywords machine learning method, automated control system, metric proximal gradient, educational process 1 1. Introduction approach to education, contributing to better learning and improved academic results [3]. Improving an automated learning management system The transparency and fairness of the educational (ACS) is an important step towards increasing the process will also benefit from the improvement of the efficiency, quality, and safety of the educational process. automated learning management system. Automating the Automating routine administrative tasks, such as creating assessment and recording of student results reduces the risk timetables, recording attendance, and generating reports, of human error and subjectivity, which helps build trust in allows you to optimize resource management and reduce the system on the part of students and their parents. The staff workload. This allows administrators to focus on more transparency of teachers’ and administrators’ actions strategic tasks, such as improving curricula and managing provided by automated systems increases the level of the quality of education [1, 2]. responsibility and openness in the educational environment Improving the quality of the educational process is one of [4]. the key goals of improving the automated control system. The Improving data security is also an important aspect. In introduction of machine learning technologies allows for more today’s cyber threat environment, it is necessary to accurate tracking and analysis of student performance, which implement modern authentication and monitoring methods helps to identify learning problems on time and provide the to protect the personal data of students and teachers [5, 6]. necessary support. Such innovations also allow for the creation This not only reduces the risk of information compromise of individualized curricula, which provide a more personalized but also ensures compliance with regulatory data protection requirements. In addition, the use of data for analysis and CPITS-II 2024: Workshop on Cybersecurity Providing in Information 0000-0003-4051-1190 (V. Zhebka); |and Telecommunication Systems II, October 26, 2024, Kyiv, Ukraine 0009-0005-3446-5994 (O. Ananchenko); ∗ Corresponding author. 0000-0003-0653-6423 (K. Osadcha); † These authors contributed equally. 0009-0007-4620-9888 (S. Zhebka); viktoria_zhebka@ukr.net (V. Zhebka); 0009-0000-7868-8341 (A. Aronov) ananchenko.oe@gmail.com (O. Ananchenko); © 2024 Copyright for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0). katheryna.osadcha@ntnu.no (K. Osadcha); szhebka@hotmail.com (S. Zhebka); webx.ghost@gmail.com (A. Aronov) CEUR Workshop ceur-ws.org ISSN 1613-0073 372 Proceedings informed decision-making helps the management of depending on the degree of deviation from the norm. With educational institutions to identify trends, predict results, the help of classification algorithms, such events can be and develop strategies for further development. automatically filtered and forwarded to the administrator Improving the ACS also ensures adaptation to modern for further investigation. In critical cases, the system can educational trends, such as distance learning, integration of automatically block access to the system from a suspicious digital resources, and the use of interactive platforms [7]. device until the circumstances are clarified. This allows educational institutions to remain relevant and The system also monitors the devices from which the competitive in the market of educational services. In user usually accesses the system. If a new device appears addition, improving the usability of the system contributes that has not been used before, the system can request to the satisfaction of students, teachers, and administration, confirmation from the user or simply inform the which is an important factor in creating a positive administrator to avoid unauthorized access. educational environment [8–15]. Integration of machine learning into ACS systems Thus, the improvement of the ACS is necessary to significantly increases the level of security and management ensure an efficient, high-quality, and safe educational efficiency. The introduction of such technologies allows for process that meets modern requirements and challenges. automated tracking of student progress, as well as timely This not only improves the management and provision of detection and response to suspicious user activity, which education but also contributes to the overall development of contributes to the overall reliability and safety of the educational institutions in a dynamic world [16–19]. educational process. Therefore, the purpose of the paper sis to improve the It is proposed to use the method of metric proximal automated learning management system by introducing an gradient as a machine learning method. advanced machine learning method [20–23]. The algorithm of the metric proximal gradient: 2. Research results 1. Select the starting point 𝑥 ∈ 𝑅 . 2. Update the metric Mn. The automated control system (ACS) implements a y n 1  x n   M n  f  x n  1 specialized database to store logs of all events occurring in the system. The logs contain information about user authentication, changes in schedules, grades, documents, x n1  proxq ,Mn  y n1  and other critical operations. Such a database provides high  1 2  detail and the ability to retrospectively analyze user actions. arg min  q  x   y n1  x n  An important aspect is the ability to sort logs by various x  2 M  parameters (user, subject, document, etc.), which allows the administrator to obtain the necessary information for y  y  n1 n 3. If 2 so, then the algorithm stops. analysis and decision-making. Otherwise, repeat step 2. Machine learning (ML) in an automated learning 4. The study of the choice of algorithm step size to management system can be effectively used to analyze ensure fast convergence revealed that the diagonal student performance. Using clustering and regression step size is more effective than the scalar step size. algorithms, the system can track the dynamics of the However, it is necessary to develop clear rules for academic performance of individual students or groups. For determining the diagonal step size in convex example, based on historical data on grades, attendance, and optimization algorithms. assignments, the system can predict the likelihood of 5. The standard proximal gradient method is successful completion of a course or identify students who guaranteed to converge in sufficiently small steps need additional support. 1 , provided that f is L-smooth. In this case, In addition, a ‘Decency’ or ‘Integrity’ rating system can  be introduced that automatically assigns ratings to students L based on their behavior (completing assignments on time, without using the Lipschitz constant, it is possible attending classes, etc.). Such ratings can serve as an to search over several rows with a return strategy. additional motivational tool for students. The methods used in this case guarantee Another important function of ML in an ACS system is convergence. to detect suspicious user activity. For this purpose, anomaly 6. A non-monotonic linear search on a string allows detection algorithms are used that can identify deviations the objective function F(x) to grow between from a typical user behavioral pattern. For example, if a user iterations but leads to a possible decrease in its who normally interacts with the ‘timetable’ and ‘grades’ value. Given the current iteration xn, the initial tabs suddenly logs in from another device and immediately metric Mn, and the potential next iteration xn + 1 accesses confidential documents such as ‘R&D’, the system check whether the following (Mn, xn + 1) criterion is generates an alarm. satisfied. The system can automatically determine the priority level (e.g., critical, high, medium, low) for each incident, 373 Figure 1: Algorithm of the metric proximal gradient  1 The next step is to perform a return, changing the scale F  x n1   F n  x n1  x n n 2 of the metric M with the coefficient   1 . n 2 M  The next step is to analyze the convergence of the where K LC  1 is the search parameter, and F k is found algorithm. Assume that the function f is L-smooth and as follows: M n  0 . The algorithm of the variable metric proximal   F n  max F  x n  , F  x n1  ,....., F x  n  min  K LC , n 1   gradient with diagonal metric will have the following steps: 1. Set the parameters, starting points, and starting metrics. 2. Calculate and by the formula. 374 Start Calculate Set parameters Set the initial points Set the initial metric Calculate Calculate Calculate Yes No Return metric Let's initialize Calculate Yes No No Stop criterion satisfied? Yes Yes No Let's get the results End Calculate Figure 2: Improved metric proximal gradient algorithm Set the parameters K ls  1,   1,   0 ,  BB n   BB  c n , y n   starting points x 0 , x1  R n , and starting metric   BB n 2 , if  BB1   BB 2 n n M S 0 n    n 1 n Calculate  BB n and  BB by the formula n   BB1   BB 2 , in other cases 1 2   Initialize Mn according to the formula 375  References  1 cin yin   min1 1 ,  n [1] V. Buriachok, et al., Implementation of Active   ББ1  cin     ББ1 n 2 Cybersecurity Education in Ukrainian Higher School,  Information Technology for Education, Science, and  1 cin yin   min1 1 Technics, vol. 178 (2023) 533–551. doi:10.1007/978-3- mi   n ,  n   ББ 2  ci    ББ 2 031-35467-0_32. n k 2 [2] V. Buriachok, V. Sokolov, Implementation of Active  n n Learning in the Master’s Program on Cybersecurity,  ci yi   mi , n 1 in other cases. Advances in Computer Science for Engineering and    i  n 2 c   Education II, vol. 938 (2020) 610-624. doi:10.1007/978- 3-030-16621-2_57. Calculate [3] V. Buhas, et al., Using Machine Learning Techniques  x n 1 :  proxq , M n x n   M  f  x n  n 1  . to Increase the Effectiveness of Cybersecurity, in: Cybersecurity Providing in Information and Telecommunication Systems, vol. 3188, no. 2 (2021) Repeat the expressions 273–281. [4] P. Skladannyi, et al., Improving the Security Policy of M n :  M n , the Distance Learning System based on the Zero Trust   Concept, in: Cybersecurity Providing in Information x n1 :  proxq ,M n x n   M n  f  x n  . 1 and Telecommunication Systems, vol. 3421 (2023) 97– 106. as long as the criterion is met [5] S. Shevchenko, et al., Protection of Information in  1 F  x n1   F n  x n1  x n n , 2 Telecommunication Medical Systems based on a Risk- Oriented Approach, in: Workshop on Cybersecurity 2 M Providing in Information and Telecommunication   F n  max F  x n  , F  x n1  ,....., F x n  min  K ls , n 1   . [6] Systems, vol. 3421 (2023) 158–167. V. Astapenya, et al., Conflict Model of Radio Return the metric Mn and perform the next iteration Engineering Systems under the Threat of Electronic xn + 1. Warfare, in: Workshop on Cybersecurity Providing in Repeat steps 2–6 until the stopping criterion is satisfied Information and Telecommunication Systems, CPITS, vol. 3654 (2024) 290–300. 3. Conclusions [7] H. Hulak, et al. Formation of Requirements for the Electronic RecordBook in Guaranteed Information The use of the metric proximal gradient method in the Systems of Distance Learning, in: Workshop on automated learning management system (ALMS) is a Cybersecurity Providing in Information and promising approach to solving optimization problems and Telecommunication Systems, CPITS 2021, vol. 2923 improving the system’s performance. This method allows (2021) 137–142. you to solve complex optimization problems where [8] V. Zhebka, et al., Optimization of Machine Learning conventional gradient methods may be ineffective due to Method to Improve the Management Efficiency of the presence of irregularities or special structures in the Heterogeneous Telecommunication Network, in: objective function. In particular, the metric proximal Cybersecurity Providing in Information and gradient method effectively copes with sparse or Telecommunication Systems, vol. 3288 (2022) 149– heterogeneous data, which is typical in large educational 155. systems. [9] P. Anakhov, et al., Systematization of Measures on Due to its ability to take into account the local geometry Lightning Protection of the Objects of of the problem, this method can be used to adaptively tune Telecommunications Network, Int. J. Adv. Trends model parameters, which is especially important in dynamic Comput. Sci. Eng. 9(5) (2020) 7870–7877. environments where conditions change over time. For [10] A. Canale, N. Lunardon, Churn Prediction in example, it can be used to optimize curricula, personalize Telecommunications Industry. A Study based on educational routes for students, or detect anomalies in user Bagging Classifiers Telecom, Carlo Alberto behavior that require immediate response. Notebooks 350 (2014) 1–11. In general, the use of the metric proximal gradient [11] DSTU (State standard of Ukraine) 3899:2013. Dyzain i method in ACS systems is recommended for tasks requiring Erhonomika. Terminy ta Vyznachennia Poniat high accuracy, adaptability, and efficient processing of [Design and Ergonomics. Terms and Definitions]. complex data structures. This will increase the level of [12] J. Hughes, M. Haran, P. C. Caragea, Autologistic automation, forecast accuracy, and safety in the Models for Binary Data on a Lattice. Environmetrics management of the educational process. 22(7) (2011) 857–871. [13] ITU-T Recommendation G.602. Transmission Media Characteristics. Reliability and Availability of Analogue Cable Transmission Systems and Associated Equipments. 376 [14] A. A. Khan, J. Sanjay, M. M. Sepehri, Applying Data Mining to Customer Churn Prediction in an Internet Service Provider. Int. J. Comput. Appl. 9(7) (2010) 8– 14. [15] O. Klymovych, et al., The Diagnostics Methods for Modern Communication Tools in the Armed Forces of Ukraine based on Neural Network Approach, Modern Machine Learning Technologies Workshop, vol. 2631 (2020) 198–208. [16] M. J. Wainwright, M. I. Jordan, Graphical Models, Exponential Families, and Variational Inference, Found. Tr. Mach. Learn. 1(1–2) (2008) 1–305. [17] P. Babarczi, et al., A Mathematical Framework for Measuring Network Flexibility, Comput. Commun. 164 (2020) 13–24. [18] V. Mukhin, et al., Models for Analysis and Prognostication of the Indicators of the Distributed Computer Systems’ Characteristics, Int. Rev. Comput. Software 10(12) (2015) 1216–1224. [19] W. Kellerer, et al., How to Measure Network Flexibility? A Proposal for Evaluating Softwarized Networks, IEEE Communications Magazine 56(10) (2018) 186–192. [20] W. Zucchini, I. L. MacDonald, R. Langrock, Hidden Markov Models for Time Series: An Introduction Using R, Chapman and Hall (2016). [21] Z. Hu, et al., Distributed Computer System Resources Control Mechanism based on Network-Centric Approach, Int. J. Intell. Syst. Appl. 9(7) (2017) 41–51. [22] V. Zhebka, Modeliuvannia Markivskoho Vypadkovoho Polia z Metoiu Yoho Podalshoi Optymizatsii ta Zastosuvannia, Scientific-Practical Journal Zviazok 5 (2020) 35–40. [23] V. Zhebka, et al., Methodology for Predicting Failures in a Smart Home based on Machine Learning Methods, in: Cybersecurity Providing in Information and Telecommunication Systems, vol. 3654 (2024) 322–332. 377