Design of an efficient VARMA GRU LSTM based predictive Trust Model for high-performance Blockchain networks

Abstract

The establishment of trust between nodes in high-performance blockchain networks remains a significant obstacle in secure applications in different fields. Herein is presented a novel predictive trust model that utilizes the power of VARMA (Vector Autoregressive Moving Average) in conjunction with GRU (Gated Recurrent Unit) and LSTM (Long Short-Term Memory) neural networks to effectively extract the trust levels of nodes based on their temporal and spatial performance metrics. The extracted features were utilized to train the VARMA Model to predicts the future trust levels of nodes. Utilizing LSTM and GRU networks under various attack scenarios, such as DDoS, Finney, Sybil, and Worm Hole show significant improvements in results. This system achieved a remarkable 10.5% reduction in communication delays, 2.5% improvement in PDR, 8.3% reduction in energy consumption, and 4.5% improvement in throughput, showing direct influences in the overall performance, security, and dependability of blockchain networks. First, by incorporation of LSTM and GRU networks, the designed system captures and analyzes the complex temporal dependencies in performance metrics, resulting in more precise predictions. Second, the integration of VARMA provides a solid basis for time series analysis, allowing for accurate forecasts of trust levels. Thirdly, this model outperforms existing trust models in multiple attack scenarios, demonstrating its resilience and efficacy in the face of adversarial actions.

URN:NBN:sciencein.jist.2024.v12.708