PBFT enhancement via grouping and machine learning-based credit evaluation

The consensus mechanism is a fundamental part of any blockchain system, crucial to ensuring security and optimal performance. Consensus mechanisms such as Practical Byzantine Fault Tolerance (PBFT) are widely adopted and have gained significant popularity. However, PBFT faces significant challenges, including high latency, high communication overhead, compromised security, and limited throughput and performance. This paper presents the Enhanced Practical Byzantine Fault Tolerance (EPBFT) consensus mechanism that integrates dynamic grouping, Boneh–Lynn–Shacham (BLS) digital signatures, and a machine learning-based trust model to address the traditional PBFT limitations. In EPBFT, nodes are divided into groups whose sizes are adaptively adjusted according to average trust values, malicious node density, and network performance indicators such as throughput and latency, ensuring a balance between scalability and fault tolerance. Within each group, master nodes are elected based on their dynamically updated trust values, selecting the node with the highest trust score. To achieve this, a machine learning-based linear regression model is employed to predict and refine node trust scores from consensus performance data, allowing malicious or unreliable nodes to be gradually excluded. BLS signatures provide lightweight, verifiable authentication for all inter-node messages, strengthening security. Simulation results in Python demonstrate that EPBFT outperforms traditional PBFT, achieving higher throughput and lower consensus overhead compared to existing consensus mechanisms. It also shows lower consensus delay compared to PBFT, and Group-based PBFT and Credit-based BFT. In general, EPBFT achieves higher scalability, stronger security, and more efficient consensus in large-scale blockchain environments.