Technological Challenges of Integrating the ISO 20022 Standard into Payment Systems Using Machine Learning

In the scope of the publication the migration of the worldwide financial industry to the ISO 20022 standard has been analyzed by considering the technical restrictions of ML algorithms applied to process a set of transactional data, the architectural restrictions of common payment systems that do not allow for the integration of smart components, and the use of smart data mapping algorithms to harmonize heterogeneous formats and fields in a single message model. Another aspect is the issue of data quality, which is necessary for reliable training of neural networks in the application areas of money laundering or fraud prevention. The objective is simplified to the construction of a theoretical model, which allows identifying opportunities to optimize the integration processes in the introduction of ISO 20022 and the subsequent use of the machine learning tooling in the transaction circuit. The method includes systems analysis and synthesis of the research results, which enables the comparison of architectural variants and the identification of conditions for adequately trading off interoperability and computational costs. The empirical-analytical part is based on publications of the last few years indexed in Scopus and Web of Science. In addition, this also ensures the topicality of the scientific basis used and the representativeness of the identified trends. The conclusion provides recommendations that reduce the risks of transition to the standard and the potential negative impact on data quality and process control. Our findings are of practical relevance to architects designing payment solutions, Data Science experts, and financial analysts.