A SUSPICIOUS FINANCIAL TRANSACTION DETECTION MODEL USING AUTOENCODER AND RISK-BASED APPROACH

cyber security and machine learning

In the modern financial ecosystem, the rapid evolution of digital technologies has not only enabled convenience and global accessibility in monetary transactions but has also opened new avenues for financial crimes. One of the most prominent and damaging forms of financial crime is money laundering — a process through which illegally obtained funds are made to appear legitimate. As criminals develop more sophisticated techniques for concealing their activities, the challenge of detecting suspicious transactions becomes increasingly complex. Conventional Anti-Money Laundering (AML) frameworks, which largely depend on manually defined rules and thresholds, are increasingly rendered insufficient due to their static nature and lack of adaptability to evolving threat landscapes. These limitations have prompted financial institutions and researchers to explore more intelligent, flexible, and data-driven approaches for monitoring financial transactions. This study introduces a novel model for detecting suspicious financial activities by integrating unsupervised deep learning techniques—particularly autoencoders—with a risk-based approach (RBA), thereby advancing the capabilities of AML systems. Traditional AML detection systems are typically built around predefined rules or statistical models that flag transactions based on threshold values or anomaly scores. Although these systems are easy to interpret and have been widely used for years, they are often unable to detect newly emerging patterns or atypical behaviors not previously encountered. Additionally, these rule-based systems tend to generate a high volume of false positives, burdening compliance officers with exhaustive investigations and potentially overlooking real threats. As a result, financial institutions are increasingly shifting toward machine learning models, which are capable of learning complex patterns in data and making more nuanced inferences.

This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.