| Abstract [eng] |
The aim of this thesis is to develop an automated transaction detection method for identifying bot activity in the Solana blockchain network. The main objectives include summarizing the types of automated transactions (bots) and their impact on blockchain systems. Creating a methodology for wallet behavior analysis and anomaly detection in the Solana network using clustering and anomaly detection algorithms. Applying K-means, DBSCAN, and Isolation Forest algorithms to identify typical and atypical wallet behavior patterns; and analyzing the results and providing recommendations for improving the security and transparency of the Solana network. The study employs a quantitative approach, utilizing real transaction data collected from the Solana blockchain via the JSON-RPC API. The data is preprocessed, normalized, and stored in an SQLite database. The methodology involves feature engineering; clustering (K-means and DBSCAN); and anomaly detection (Isolation Forest), implemented using Python libraries such as Pandas, NumPy, and Scikit-learn. The effectiveness of the methods is evaluated using metrics such as the Silhouette coefficient, Calinski–Harabasz index, Precision, Recall, and F1-score. The experimental results, based on the analysis of 100 Solana network wallets, demonstrate the ability of the proposed method to identify automated transaction patterns (including MEV, “sandwich,” and frequency trading bots) and detect anomalies. In particular, the combination of DBSCAN and Isolation Forest proves to be the most effective, achieving a favorable balance between Precision, Recall, and F1-score. The main conclusions highlight the importance of adapting bot detection methods to the specific characteristics of the Solana network and the need for regular model updates to keep pace with the evolving nature of bot activity. The thesis also underlines the potential of the proposed methodology for real-time integration into network security systems and DeFi platforms, strengthening both security and market transparency. The work consists of 86 pages of text (excluding tables, figures, list of abbreviations, references, and appendices), 24 figures, and 28 tables. |
