PREDICTING FRAUDULENT CRYPTOCURRENCY TRANSACTIONS USING LOGISTIC REGRESSION

Abstract

The article conducts a SWOT analysis of cyber threats, identifies the
strengths and weaknesses of cryptocurrencies. It is determined that the development
of cryptocurrencies forms a new class of digital assets, which is attracting increasing
attention from the economic and financial communities and information technology.
The issue of detecting fraudulent transactions with cryptocurrency is highlighted. To
solve the problems of detecting fraudulent transactions, the authors propose to use new
technologies based on data analysis methods, in particular, the development of logistic
regression models. The following algorithm for classifying fraudulent transactions with
cryptocurrency is proposed, which is reduced to the classical data classification scheme.
The following steps are highlighted: data loading into the dataset and primary analysis,
data preparation for analysis, division into training and test samples, application of the
classification algorithm on the training sample, evaluation of the model accuracy on
the test sample, model optimization if necessary, and conclusion, where, if the model
accuracy is high, it can be used to classify fraudulent transactions. The main task of the
presented stages is to detect suspicious cryptocurrency transactions with as few false
positives as possible. The classification of cryptocurrency transactions is proposed to
be carried out with the Ethereum cryptocurrency. The R language and its integrated
processing environment R Studio are chosen as tools. A logistic regression model has
been developed to detect fraudulent transactions with cryptoassets. The model checks
a new transaction for fraud. The model's high accuracy of 98 percent demonstrates its
effectiveness. The model can be improved to take into account new types of fraudulent