Application of machine learning methods for detecting attacks on blockchain systems

The subject of the research is methods for detecting attacks in networks with the Proof-of-Stake (PoS) consensus mechanism. The purpose of this experimental investigation and analysis is to evaluate the effectiveness of classical machine learning algorithms for detecting malicious nodes in blockchain systems. The tasks include the analysis of blockchain technology vulnerabilities, the creation and use of a specialized dataset for PoS networks, as well as the construction and testing of machine learning models. The main focus is placed on comparing three algorithms – Random Forest, Support Vector Machine, and k-Nearest Neighbors – in order to determine their suitability for monitoring node activity and detecting anomalies. To solve the tasks set, the following methods were implemented: modeling, empirical, and mathematical approaches were applied. Modeling consisted of software implementation of the selected algorithms and subsequent analysis of their performance using accuracy, recall, F1-score metrics, and confusion matrices. Empirical methods were realized through testing the models on a partially synthetic dataset containing more than 10,000 records of blockchain nodes and transactions. Mathematical methods involved the calculation of statistical performance indicators and the analysis of feature importance that characterizes node behavior. The achieved results include the validation of a dataset for PoS blockchains that incorporates key operational parameters of transactions and nodes, the development of recommendations for further use of machine learning models, and the testing of selected models. Conclusions. The study demonstrated that machine learning is an effective tool for identifying anomalies and malicious activity in blockchain systems. The obtained results lay the foundation for further research, which may focus on expanding the feature space, integrating deep neural networks, developing ensemble approaches, and adapting methods to different types of blockchains.