SecureLedger: A Hybrid Blockchain–Machine Learning Framework for Real-Time Fraud Detection and Tamper-Evident Financial Transaction Recording

Abstract

There is a rise in losses resulting from transaction fraud in financial institutions; Decentralized Finance (Defi) platforms have suffered about $10.1 billion worth of loss during 31 major attacks from 2020 through 2023 [3]. In this paper, SecureLedger is introduced, which is a working Flask framework integrating a Random Forest classifier for fraud prediction together with a custom-designed SHA-256 chained hash ledger such that each transaction is fraud scored and permanently written to ledger as a block. With respect to a synthetic dataset of 2,000 transactions, the model achieved 99.67%, 98.61%, and 0.9924 in terms of accuracy, precision/recall/F1-score, and Area Under the Curve (AUC), respectively, where account age and transaction velocity were found to be the main predictors. The live admin dashboard and chain integrity check functionality in the running prototype are validated to show the true state of the ledger as well as detecting any intentional tampering. Compared to other six recent works in the application of blockchain technology in financial security, SecureLedger gives up architecture complexity for clarity and reproducibility.