Transforming Payment Systems Through AI And ML: A Cloud-Native Approach

Abstract

Payment systems, powered with AI and ML, entail smart routing of payment transactions. Payment systems, which handle billions of dollars in e-commerce transactions all over the world, are fraught with failures. These failures leave the customers disappointed and the payment providers with enormous losses. This work studies a payment transaction routing problem to improve the success rate for the payments. A large payment processor handles millions of payment transactions, which are routed to a specific terminal and processed further. However, this process is fraught with failures, which leaves the customers disappointed and the payment channel operators (PCOs) with enormous losses. The success rate for the payment transactions is computed using routing metalearning and cloud-native architecture. The architecture of the payment systems in which the pipelines are to be integrated is presented, along with the high-level components of such systems. A real-world payment dataset is presented, which contains timestamped records of all the payment transactions processed during a week in a mid-sized European country. It is well known that understandability, simplicity, and scalability are some of the great boons of early machine learning (ML) models/families. Data drift is a very dangerous problem in ML models, which refers to changes in the training data distribution that can lead to a decrease in the model performance . For some time, sophisticated models with higher predictive powers overshadowed the early models. But it later turned out that the black box nature of the sophisticated models renders them more vulnerable, especially in cases like payment systems, where an explanation is of utter importance .