A Machine Learning-Based Routing Technique for Off-chain Transactions in Payment Channel Networks

Abstract

Blockchain is a foundational technology that has the potential to create new prospects for our economic and social systems. However, the lack of scalability limits the capability to deliver a target throughput and latency compared to the traditional financial systems. Layer-two is a collective term for solutions designed to enhance scalability by handling transactions off the main chain. For example, bidirectional payment channels allow the execution of fast transactions between two parties, thus forming the so-called payment channel networks (PCNs). Therefore, efficient routing protocols are needed to find a payment path from the sender to the receiver, with the lowest transaction fees. These protocols have to consider, among other factors, the unexpected online/offline behavior of the constituent payment nodes and the payment channel imbalance. This study proposes a novel machine learning-based technique for off-chain transaction routing. It is a fully distributed approach to be used within PCNs. For this purpose, the effect of the offline nodes and channel imbalance on the payment channels network are modeled. The simulation results demonstrate that the proposed technique strikes a good balance when considering the success ratio, transaction fees, routing efficiency, transaction overhead, and transaction maintenance overhead compared to other techniques that have previously been proposed for the same purpose.