NPCI Gateway Integration for Enhanced Fraud Prevention in Digital Transactions (Journal of Research in Science Computing and Engineering)

Summary

• The platform achieved a fraud class precision of 94.2% using a hybrid machine learning approach with Random Forest and Isolation Forest classifiers.p.4
• The system reached a recall rate of 91.5% and an F1-Score of 92.8% for fraudulent transaction identification.p.4
• Average transaction processing time was measured at 850 ms, while fraud flagging latency was 120 ms post-transaction.p.4
• The platform maintained a peak throughput of 95 transactions per second during load testing.p.4
• Transaction logs hashed with SHA-256 required 1.5 ms per entry and registered a 0% collision incidence.p.4
• Users experienced an average completion time of 1.3 seconds per transaction, yielding an 89% satisfaction score.p.4
• Testing recorded 0 downtime incidents while evaluating system availability and dynamic load balancing.p.4

Key insights

1. Integrating Random Forest and Isolation Forest classifiers produced an AUC-ROC score of 0.96 in evaluating fraud detection capabilities.p.4
2. Asynchronous API calls and edge device caching successfully reduced server load during peak performance testing.p.4
3. Fraud flagging latency was minimized to 120 ms post-transaction, allowing near-instantaneous anomaly detection.p.4
4. Cryptographic linkage of logs via SHA-256 hashing ensured tamper-proof records with zero recorded collision instances.p.4
5. Dynamic load balancing techniques including Round Robin and Least Connections effectively prevented service degradation under high transaction loads.p.4
6. Integrating synthetic data with sandbox APIs provided an effective strategy for overcoming class imbalance in fraud detection model training.p.4