Bayesian Optimization for a Permissioned Blockchain: A Data-Driven Approach to Optimal Parameter Selection

Optimizing the performance of permissioned blockchain networks like Hyperledger Fabric remains a critical challenge due to the complexity and interdependence of system-level parameters. This study presents a data-driven approach using Bayesian Optimization (BO) to minimize transaction latency by tuning key parameters-block size, block composition time, and transactions per block. Employing surrogate models including Random Forest Regressor, Gradient Boosting Regressor, Gaussian Process Regressor, and direct dataset lookup, the researchers evaluated 71 configurations to identify optimal settings. Results show that BO significantly reduced latency compared to default configurations, with the direct dataset lookup method achieving the best performance at 114 ms. The study also highlights the convergence behavior of different models and the interdependencies among parameters. This work forms part of a broader effort to deploy an optimized, blockchain-based web application for educational credential management. Future research will incorporate resource utilization metrics and explore adaptive optimization in dynamic environments.