Study of the possibility of integrating blockchain technologies and ML models into financial risk management systems

Abstract

The article examines the feasibility of developing a model of an information technology based on the integration of the Hyperledger Fabric blockchain technology and ML models for financial risk management. The relevance of the study is driven by the growing volume of transactions in modern financial institutions, the increasing complexity of financial instruments, and the introduction of stricter regulatory requirements, which intensify the need to ensure data immutability and authenticity.

The study analyzes the characteristics of traditional centralized risk management systems, particularly their limitations in ensuring the immutability of historical data, which creates potential risks of manipulation and distortion of analytical results. Since ML models, widely used for credit scoring and fraud detection, critically depend on the quality of input data, there arises a need to address a scientific and applied problem—namely, the creation of an integrated architecture that combines blockchain-based guarantees of data immutability with the predictive power of ML models. In preparing the article, an analysis of international and domestic publications was conducted regarding the application of machine learning algorithms (logistic regression, Random Forest, XGBoost, Isolation Forest) in financial scoring, as well as the effectiveness of permissioned blockchain architectures, particularly Hyperledger Fabric.

It is substantiated that the integration of a blockchain layer as a trust infrastructure for ML processes (provenance, audit trail, integrity of updates, governance) remains insufficiently formalized. A three-layer architecture (Data Layer, Analytics Layer, Governance Layer) is proposed, along with formal definitions of the transaction space, ML-based risk model, and the integrated risk formula.

During the study, a synthetic dataset was used, and a comparison of the performance of Logistic Regression, Random Forest, and XGBoost models was conducted.
The modeling results show that the XGBoost model achieved the highest performance, while integration with blockchain did not affect predictive efficiency but ensured full traceability of decisions and improved audit transparency. The results are of a predictive nature and require further empirical validation under real-world implementation conditions in financial institutions.