FAMMNPB: Design of an Enhanced Forensic Analysis Model with Multidomain Feature Extraction and Advanced Neural Architecture for Permissioned Blockchains

This work introduces a novel approach that leverages multidomain feature extraction coupled with a 1D Convolutional Neural Network (CNN) for robust anomaly detection. To address the challenge of transaction delay predictions, we employ a Bi-directional Gated Recurrent Unit (BiGRU) RNN, for its capability to capture temporal sequences with high precision. For the task of chain splitting, the Teaching-Learning-based Optimization (TLbO) technique is used to ensure the efficient creation of sidechains. Further, we integrated DeepSHAP, in tandem with 1D CNN and BiGRU, to enhance the performance of our models, fostering better trust in the results. Our proposed methodology yields notable improvements over existing techniques. As we observed enhancements in the precision of anomaly detection by 4.9%, accuracy in detecting anomalous events by 3.5%, Recall by 4.5% and AUC by 3.9%. These advancements fortify the security posture of permissioned blockchains and optimize their operational efficacy, thus marking a significant improvement in blockchain forensics.