Кафедри
Постійне посилання на розділhttps://repository.kpi.kharkov.ua/handle/KhPI-Press/35393
Переглянути
7 результатів
Результати пошуку
Документ Machine learning(National technical university "Kharkiv polytechnic institute", 2024) Gavrylenko, SvitlanaThe workshops guide contains the necessary material for performing workshops: options for tasks, examples of program texts and report. Intended for students of computer majors at higher educational institutions.Документ Identification of the state of an object under conditions of fuzzy input data(ПП "Технологічний Центр", 2019) Semenov, Serhii. G.; Sira, Oksana; Gavrylenko, Svitlana; Kuchuk, Nina G.Проведена модернізація методів ідентифікації стану об'єктів в умовах нечітких вхідних даних, описаних своїми функціями належності. Обраний напрямок вдосконалення традиційних методів пов'язаний із принциповими особливостями вирішення цього завдання в реальних умовах малої вибірки вхідних даних. У цих умовах для розв’язання задачі ідентифікації стану доцільно перейти до менш вибагливої в інформаційному відношенні технології опису вихідних даних, заснованої на математичному апараті нечіткої математики. Цей перехід зажадав розробки нових формальних методів вирішення конкретних завдань. При цьому для багатовимірного дискримінантного аналізу розроблено методику розв’язання нечіткої системи лінійних алгебраїчних рівнянь. Для вирішення завдання кластеризації запропонована спеціальна процедура порівняння нечітких відстаней між об'єктами кластеризації і центрами групування. Обраний напрямок вдосконалення традиційного методу регресійного аналізу визначено неможливістю використання класичного методу найменших квадратів в умовах, коли всі змінні описані нечітко. Ця обставина привела до необхідності побудови спеціальної двохкрокової процедури вирішення завдання. При цьому реалізується мінімізація лінійної комбінації міри видалення шуканого рішення від модального і міри компактності функції приналежності пояснювальної змінної. Технологія нечіткого регресійного аналізу реалізована в важливому для практики випадку, коли вихідні нечіткі дані описані загальними функціями приналежності (L-R) типу. При цьому отримано аналітичний розв'язок задачі у вигляді розрахункових формул. В результаті обговорення показано, що модернізація класичних методів рішення задачі ідентифікації стану з урахуванням нечіткого характеру представлення вихідних даних дозволила проводити ідентифікацію об'єктів в реальних умовах малої вибірки нечітких вихідних даних.Документ Development and comparative analysis of computer system state identification methods based on ensemble algorithms(Інжиніринг, 2020) Gavrylenko, Svitlana; Sheverdin, IlliaThe scientific novelty of the results obtained consists in creating ensemble methods for classifying the state of a computer system without a teacher and with a teacher. The method based on the "Isolation Forest" algorithm can be used as an express method for analyzing a computer system state. This will allow not only to identify the state of a computer system state, but also to highlight the name of the abnormal processes. This method can also be used to generate labeled data and use it as the source data of the ensemble algorithm with a teacher. The algorithm with a teacher built according to the C4.5 algorithm is more accurate and can be used to refine the result of identifying a computer system state using the method based on the "Isolation Forest" algorithm.Документ Method of computer system state identification based on boosting ensemble with special preprocessing procedure(Національний технічний університет "Харківський політехнічний інститут", 2022) Chelak, Viktor; Gavrylenko, SvitlanaThe subject of the research is methods of identifying the state of the Computer System. The object of research is the process of identifying the state of a computer system for information protection. The aim of the research is to develop the method for identifying the state of a computer system for information protection. This article is devoted to the development of method (boosting ensemble) to increase the accuracy of detecting anomalies in computer systems. Methods used: artificial intelligence methods, machine learning, decision tree methods, ensemble methods. The results were obtained: a method of computer system identification based on boosting ensemble with special preprocessing procedure is developed. The effectiveness of using machine learning technology to identify the state of a computer system has been studied. Experimental researches have confirmed the effectiveness of the proposed method, which makes it possible to recommend it for practical use in order to improve the accuracy of identifying the state of the computer system. Conclusions. According to the results of the research, ensemble classifier of computer system state identification based on boosting was proposed. It was found that the use of the proposed classifier makes it possible to reduce the variance to 10%. In addition, due to the optimization of the initial data, the efficiency of identifying the state of the computer was increased. Prospects for further research may be to develop an ensemble of fuzzy decision trees based on the proposed method, optimizing their software implementation.Документ Development of a method for identifying the state of a computer systemusing fuzzy cluster analysis(Національний технічний університет "Харківський політехнічний інститут", 2020) Gavrylenko, Svitlana; Chelak, Viktor; Hornostal, Oleksii; Vassilev, VelizarThe subject of this article is the study of methods for identifying the state of computer systems. The purpose of the article is to develop a method for identifying the abnormal state of a computer system based on fuzzy cluster analysis. Objective: to analyze methods for identifying the state of computer systems; to conduct research on the selection of source data; to develop a method for identifying the state of a computer system with a small sample or fuzzy source data; to investigate and justify the procedure for comparing fuzzy distances between grouping centers and clustering objects; to develop a software and test. The methodsused in the paper: cluster analysis, fuzzy logic tools. The following resultswere obtained: a method was theoretically substantiated and investigated for identifying the state of a computer system with a small sample or fuzziness of the initial data, which is distinguished by the use of the method based on fuzzy cluster analysis by the refined grouping procedure. To solve the clustering problem, we used a special procedure for comparing fuzzy distances between grouping centers and clustering objects. Software was developed and testing of the developed method was performed. The quality of classification based on the ROC analysis is assessed. Conclusions. The scientific novelty of the results is as follows: a study was conducted on the selection of source data for analysis; a method for identifying the state of a computer system based on fuzzy cluster analysis using a special procedure for comparing fuzzy distances between grouping centers and clustering objects has been developed. This allowed to improve the classification quality to 22 %.Документ The ensemble method development of classification of the computer system state based on decisions trees(Національний технічний університет "Харківський політехнічний інститут", 2020) Gavrylenko, Svitlana; Sheverdin, Illia; Kazarinov, MichaelThe subject of this article is exploration of methods for identifying the status of a computer system.The purpose of the article is development of a method for classifying a computer system anomalous state based on ensemble methods. Task: To investigate the usage of algorithms for building decision trees: REPTree, Random Tree, J48, HoeffdingTree, DecisionStump and bagging and boosting decision tree ensembles to identify a computer system anomalous state by analyzing operating system events. The methods used are artificial intelligence, machine learning and ensemble classification methods. The following results were obtained: the methods of identifying the computer systems anomalous state based on ensemble methods were investigated, namely, bagging, boosting, and classifiers: REPTree, Random Tree, J48, HoeffdingTree, DecisionStump to identify a computer system anomalous state. The different classifiers set and classifiers ensembles were developed. Training and cross-validation on each algorithm was performed. The developed classifiers performance has been evaluated. The research suggests an ensemble method ofa computer system state classifying based on the J48 decision tree algorithm. Conclusions.The scientific novelty of the obtained results consists in creating an ensemble method for classifying the state of a computer system based on a decision tree, which makes it possible to increase the reliability and speed of classification.Документ Development of anomalous computer behavior detection method based on probabilistic automaton(National University of Civil Protection of Ukraine, 2019) Chelak, Viktor; Chelak, E.; Gavrylenko, Svitlana; Semenov, SerhiiThis work proposes anomalous computer system behavior detection method based on probabilistic automaton. Main components of the method are automaton structure generation model and its modification procedure. The distinctive feature of the method is the adaptation of the automaton structure generation procedure for detecting attack scenarios of the same type, by restructuring the automaton upon a match and by recalculating the probability of state changes. Proposed method allows to speed up the detection of anomalous computer behavior, as well as to detect anomalies in computer systems, scenario profiles of which only partially match the instances used to generate automaton structure. The obtained results allow us to conclude that the developed meth-od can be used in heuristic analyzers of anomaly detection systems.