Abstract: Leveraging a comprehensive dataset from DNB, Norway's largest bank, this study aims to design, detail, and assess a machine learning system tailored to prioritize financial transactions for manual review in the context of potential money laundering. The model employs supervised machine learning techniques and draws on three categories of historical data: transactions flagged as suspicious by the bank's internal alert system, routine legal transactions, and potential money laundering cases reported to authorities. By analyzing sender and recipient background information, historical behavior, and transaction history, the model is trained to predict the likelihood that a new transaction should be reported. The findings indicate that excluding unreported alarms and uninvestigated transactions from the training process can lead to suboptimal model performance.

Keywords: Supervised Learning, Machine Learning, Beneish Score, Hybrid Model, Suspicious Transactions, Financial Statement Fraud, Hidden Markov Model.

| DOI: 10.17148/IARJSET.2024.11702