Implementation of Machine Learning for Waqif Honesty Detection in Blockchain-Based Digital Waqf Systems (Computer Journal)

Summary

• The study integrated the Ethereum Fraud Detection Dataset containing 9,841 transactions and 51 attributes with digital waqf system behavioral data.p.1
• Data preprocessing involved cleaning, missing value handling, categorical encoding, StandardScaler normalization, and Isolation Forest anomaly detection.p.1
• The Isolation Forest model was configured with 300 n_estimators and a random_state parameter of 42 off-chain.p.4
• Transactions with a trust score of minimal 80 are assigned an Approve status and recorded on the blockchain.p.4
• Transactions in the 50–79 trust score range are assigned a Review status for manual nazhir verification.p.4
• Transactions with a trust score below 50 are assigned a Reject status and stopped prior to blockchain recording.p.4
• Representative testing generated trust scores of 92 for normal transactions, 68 for review transactions, and 27 for fraud transactions.p.1

Key insights

1. Executing machine learning inferencing off-chain prevents imposing heavy computational costs on smart contracts while establishing an effective input control layer.p.1
2. Blockchain guarantees post-recording data integrity but cannot automatically ensure the credibility of data before it is recorded.p.2
3. Smart contracts execute deterministic rules based on received trust score parameters rather than detecting novel anomaly patterns adaptively.p.2
4. The web application layer automatically records behavioral data, including access time, IP address, device fingerprints, and form input duration.p.3
5. Converting anomaly scores to a 0–100 trust score provides an interpretable operational bridge between statistical models and smart contract decision logic.p.4
6. Representative sample transaction TX20260001 yielded an anomaly score of -0,31 and a trust score of 92, resulting in a Normal prediction.p.7
7. Representative sample transaction TX20260002 produced an anomaly score of 0,12 and a trust score of 68, resulting in a Review prediction.p.7
8. Representative sample transaction TX20260003 resulted in an anomaly score of 0,45 and a trust score of 27, resulting in a Fraud prediction.p.7