PROTOCOL FOR MANAGING SECURE TRANSACTIONS IN A BLOCKCHAIN NETWORK WITH RISK-BASED ADMISSION

Abstract

The paper proposes a protocol for managing secure transactions in a blockchain network with risk-based admission, which provides adaptive pre-consensus processing without modifying the underlying consensus mechanism. The protocol relies on an off-chain risk assessment service that generates a feature profile for each transaction, calculates a risk score using a transformer-type neural network module for tabular features, and estimates the prediction uncertainty. The risk score and uncertainty are transformed by a zoning policy into White, Gray, and Black zones with the formation of a protocol directive that defines the processing mode, priority, admission rules, and audit. Three modes are introduced: STD as standard verification, ENH as enhanced verification with additional control procedures, QUAR as quarantine processing with access restrictions and logging. Experimental verification was performed on the labeled part of the graph transaction set, in which, after removing unlabeled examples, 46,564 transactions with a profile of 188 features and a partition into training, validation and test subsamples were used; class imbalance was taken into account by weight correction of the positive class. The test achieved ROC-AUC 0.9207 and AUPRC 0.7588, which confirms the suitability of the risk assessment for protocol control The configured zoning thresholds formed a flow distribution of 0.6845 in White, 0.2929 in Gray and 0.0226 in Black, ensuring high quarantine selectivity with Precision 0.9655 and escalation of risky transactions to increased control with Recall 0.8596 for Gray and Black. The microbenchmark in batch execution B=256 showed average latencies of 0.0778 ms for STD, 0.5215 ms for ENH and 0.5815 ms for QUAR, making continuous hard verification resource-intensive. With actual zone shares, the average normalized cost of pre-consensus processing is 2.816 per transaction, and the savings compared to the continuous ENH policy reach 57.98%, while maintaining control of the risk flow without interfering with the consensus