Frequent Pattern Mining for Early Detection of Fraudulent Transactions in Real-Time Digital Banking Systems

Existing digital banking fraud detection systems often rely on individual transaction attributes, which are ineffective against coordinated, repeat fraud. A single valid transaction may initiate fraud, followed by subsequent transactions that may be legitimate on their own yet fit a pattern viewed alongside frequency, timing, transaction size, and device, card, and address combinations. The work can lead to missed fraud, financial losses, and false alarms. To address the challenge, the work proposes a novel Frequent Pattern–Driven Fraud Intelligence (FP-FDI) framework for real-time digital banking systems that detects fraudulent transactions early. The proposed framework performs the following steps: transaction preprocessing, behavior pattern construction, frequent pattern extraction, suspicious pattern scoring, and fraud classification. Machine learning discovers relationships between transaction attributes, and frequent pattern mining uncovers unknown or hidden fraudulent behaviors. It provides scalable, easy-to-understand fraud analysis that is efficient for continuous transaction monitoring. The IEEE-CIS Fraud Detection dataset is publicly available. The proposed FP-FDI framework achieved 97.4% accuracy, 96.8% precision, 95.9% recall, and a 96.3% F1 score. The results show it effectively detects complex fraudulent patterns. The proposed solution can enable early fraud warnings, minimize false alarms and unnoticed fraudulent transactions, reduce financial exposure, and increase the security and integrity of real-time digital banking.